2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
37 /* control flags for do_chunk_alloc's force field
38 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
39 * if we really need one.
41 * CHUNK_ALLOC_FORCE means it must try to allocate one
43 * CHUNK_ALLOC_LIMITED means to only try and allocate one
44 * if we have very few chunks already allocated. This is
45 * used as part of the clustering code to help make sure
46 * we have a good pool of storage to cluster in, without
47 * filling the FS with empty chunks
51 CHUNK_ALLOC_NO_FORCE = 0,
52 CHUNK_ALLOC_FORCE = 1,
53 CHUNK_ALLOC_LIMITED = 2,
57 * Control how reservations are dealt with.
59 * RESERVE_FREE - freeing a reservation.
60 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
62 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
63 * bytes_may_use as the ENOSPC accounting is done elsewhere
68 RESERVE_ALLOC_NO_ACCOUNT = 2,
71 static int update_block_group(struct btrfs_trans_handle *trans,
72 struct btrfs_root *root,
73 u64 bytenr, u64 num_bytes, int alloc);
74 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
76 u64 bytenr, u64 num_bytes, u64 parent,
77 u64 root_objectid, u64 owner_objectid,
78 u64 owner_offset, int refs_to_drop,
79 struct btrfs_delayed_extent_op *extra_op);
80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
81 struct extent_buffer *leaf,
82 struct btrfs_extent_item *ei);
83 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
84 struct btrfs_root *root,
85 u64 parent, u64 root_objectid,
86 u64 flags, u64 owner, u64 offset,
87 struct btrfs_key *ins, int ref_mod);
88 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, struct btrfs_disk_key *key,
92 int level, struct btrfs_key *ins);
93 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
94 struct btrfs_root *extent_root, u64 alloc_bytes,
95 u64 flags, int force);
96 static int find_next_key(struct btrfs_path *path, int level,
97 struct btrfs_key *key);
98 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
99 int dump_block_groups);
100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
101 u64 num_bytes, int reserve);
104 block_group_cache_done(struct btrfs_block_group_cache *cache)
107 return cache->cached == BTRFS_CACHE_FINISHED;
110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
112 return (cache->flags & bits) == bits;
115 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
117 atomic_inc(&cache->count);
120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
122 if (atomic_dec_and_test(&cache->count)) {
123 WARN_ON(cache->pinned > 0);
124 WARN_ON(cache->reserved > 0);
125 kfree(cache->free_space_ctl);
131 * this adds the block group to the fs_info rb tree for the block group
134 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
135 struct btrfs_block_group_cache *block_group)
138 struct rb_node *parent = NULL;
139 struct btrfs_block_group_cache *cache;
141 spin_lock(&info->block_group_cache_lock);
142 p = &info->block_group_cache_tree.rb_node;
146 cache = rb_entry(parent, struct btrfs_block_group_cache,
148 if (block_group->key.objectid < cache->key.objectid) {
150 } else if (block_group->key.objectid > cache->key.objectid) {
153 spin_unlock(&info->block_group_cache_lock);
158 rb_link_node(&block_group->cache_node, parent, p);
159 rb_insert_color(&block_group->cache_node,
160 &info->block_group_cache_tree);
161 spin_unlock(&info->block_group_cache_lock);
167 * This will return the block group at or after bytenr if contains is 0, else
168 * it will return the block group that contains the bytenr
170 static struct btrfs_block_group_cache *
171 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
174 struct btrfs_block_group_cache *cache, *ret = NULL;
178 spin_lock(&info->block_group_cache_lock);
179 n = info->block_group_cache_tree.rb_node;
182 cache = rb_entry(n, struct btrfs_block_group_cache,
184 end = cache->key.objectid + cache->key.offset - 1;
185 start = cache->key.objectid;
187 if (bytenr < start) {
188 if (!contains && (!ret || start < ret->key.objectid))
191 } else if (bytenr > start) {
192 if (contains && bytenr <= end) {
203 btrfs_get_block_group(ret);
204 spin_unlock(&info->block_group_cache_lock);
209 static int add_excluded_extent(struct btrfs_root *root,
210 u64 start, u64 num_bytes)
212 u64 end = start + num_bytes - 1;
213 set_extent_bits(&root->fs_info->freed_extents[0],
214 start, end, EXTENT_UPTODATE, GFP_NOFS);
215 set_extent_bits(&root->fs_info->freed_extents[1],
216 start, end, EXTENT_UPTODATE, GFP_NOFS);
220 static void free_excluded_extents(struct btrfs_root *root,
221 struct btrfs_block_group_cache *cache)
225 start = cache->key.objectid;
226 end = start + cache->key.offset - 1;
228 clear_extent_bits(&root->fs_info->freed_extents[0],
229 start, end, EXTENT_UPTODATE, GFP_NOFS);
230 clear_extent_bits(&root->fs_info->freed_extents[1],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 static int exclude_super_stripes(struct btrfs_root *root,
235 struct btrfs_block_group_cache *cache)
242 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
243 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
244 cache->bytes_super += stripe_len;
245 ret = add_excluded_extent(root, cache->key.objectid,
250 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
251 bytenr = btrfs_sb_offset(i);
252 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
253 cache->key.objectid, bytenr,
254 0, &logical, &nr, &stripe_len);
258 cache->bytes_super += stripe_len;
259 ret = add_excluded_extent(root, logical[nr],
269 static struct btrfs_caching_control *
270 get_caching_control(struct btrfs_block_group_cache *cache)
272 struct btrfs_caching_control *ctl;
274 spin_lock(&cache->lock);
275 if (cache->cached != BTRFS_CACHE_STARTED) {
276 spin_unlock(&cache->lock);
280 /* We're loading it the fast way, so we don't have a caching_ctl. */
281 if (!cache->caching_ctl) {
282 spin_unlock(&cache->lock);
286 ctl = cache->caching_ctl;
287 atomic_inc(&ctl->count);
288 spin_unlock(&cache->lock);
292 static void put_caching_control(struct btrfs_caching_control *ctl)
294 if (atomic_dec_and_test(&ctl->count))
299 * this is only called by cache_block_group, since we could have freed extents
300 * we need to check the pinned_extents for any extents that can't be used yet
301 * since their free space will be released as soon as the transaction commits.
303 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
304 struct btrfs_fs_info *info, u64 start, u64 end)
306 u64 extent_start, extent_end, size, total_added = 0;
309 while (start < end) {
310 ret = find_first_extent_bit(info->pinned_extents, start,
311 &extent_start, &extent_end,
312 EXTENT_DIRTY | EXTENT_UPTODATE);
316 if (extent_start <= start) {
317 start = extent_end + 1;
318 } else if (extent_start > start && extent_start < end) {
319 size = extent_start - start;
321 ret = btrfs_add_free_space(block_group, start,
324 start = extent_end + 1;
333 ret = btrfs_add_free_space(block_group, start, size);
340 static noinline void caching_thread(struct btrfs_work *work)
342 struct btrfs_block_group_cache *block_group;
343 struct btrfs_fs_info *fs_info;
344 struct btrfs_caching_control *caching_ctl;
345 struct btrfs_root *extent_root;
346 struct btrfs_path *path;
347 struct extent_buffer *leaf;
348 struct btrfs_key key;
354 caching_ctl = container_of(work, struct btrfs_caching_control, work);
355 block_group = caching_ctl->block_group;
356 fs_info = block_group->fs_info;
357 extent_root = fs_info->extent_root;
359 path = btrfs_alloc_path();
363 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
366 * We don't want to deadlock with somebody trying to allocate a new
367 * extent for the extent root while also trying to search the extent
368 * root to add free space. So we skip locking and search the commit
369 * root, since its read-only
371 path->skip_locking = 1;
372 path->search_commit_root = 1;
377 key.type = BTRFS_EXTENT_ITEM_KEY;
379 mutex_lock(&caching_ctl->mutex);
380 /* need to make sure the commit_root doesn't disappear */
381 down_read(&fs_info->extent_commit_sem);
383 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
387 leaf = path->nodes[0];
388 nritems = btrfs_header_nritems(leaf);
391 if (btrfs_fs_closing(fs_info) > 1) {
396 if (path->slots[0] < nritems) {
397 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
399 ret = find_next_key(path, 0, &key);
403 if (need_resched() ||
404 btrfs_next_leaf(extent_root, path)) {
405 caching_ctl->progress = last;
406 btrfs_release_path(path);
407 up_read(&fs_info->extent_commit_sem);
408 mutex_unlock(&caching_ctl->mutex);
412 leaf = path->nodes[0];
413 nritems = btrfs_header_nritems(leaf);
417 if (key.objectid < block_group->key.objectid) {
422 if (key.objectid >= block_group->key.objectid +
423 block_group->key.offset)
426 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
427 total_found += add_new_free_space(block_group,
430 last = key.objectid + key.offset;
432 if (total_found > (1024 * 1024 * 2)) {
434 wake_up(&caching_ctl->wait);
441 total_found += add_new_free_space(block_group, fs_info, last,
442 block_group->key.objectid +
443 block_group->key.offset);
444 caching_ctl->progress = (u64)-1;
446 spin_lock(&block_group->lock);
447 block_group->caching_ctl = NULL;
448 block_group->cached = BTRFS_CACHE_FINISHED;
449 spin_unlock(&block_group->lock);
452 btrfs_free_path(path);
453 up_read(&fs_info->extent_commit_sem);
455 free_excluded_extents(extent_root, block_group);
457 mutex_unlock(&caching_ctl->mutex);
459 wake_up(&caching_ctl->wait);
461 put_caching_control(caching_ctl);
462 btrfs_put_block_group(block_group);
465 static int cache_block_group(struct btrfs_block_group_cache *cache,
466 struct btrfs_trans_handle *trans,
467 struct btrfs_root *root,
471 struct btrfs_fs_info *fs_info = cache->fs_info;
472 struct btrfs_caching_control *caching_ctl;
475 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
476 BUG_ON(!caching_ctl);
478 INIT_LIST_HEAD(&caching_ctl->list);
479 mutex_init(&caching_ctl->mutex);
480 init_waitqueue_head(&caching_ctl->wait);
481 caching_ctl->block_group = cache;
482 caching_ctl->progress = cache->key.objectid;
483 atomic_set(&caching_ctl->count, 1);
484 caching_ctl->work.func = caching_thread;
486 spin_lock(&cache->lock);
488 * This should be a rare occasion, but this could happen I think in the
489 * case where one thread starts to load the space cache info, and then
490 * some other thread starts a transaction commit which tries to do an
491 * allocation while the other thread is still loading the space cache
492 * info. The previous loop should have kept us from choosing this block
493 * group, but if we've moved to the state where we will wait on caching
494 * block groups we need to first check if we're doing a fast load here,
495 * so we can wait for it to finish, otherwise we could end up allocating
496 * from a block group who's cache gets evicted for one reason or
499 while (cache->cached == BTRFS_CACHE_FAST) {
500 struct btrfs_caching_control *ctl;
502 ctl = cache->caching_ctl;
503 atomic_inc(&ctl->count);
504 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
505 spin_unlock(&cache->lock);
509 finish_wait(&ctl->wait, &wait);
510 put_caching_control(ctl);
511 spin_lock(&cache->lock);
514 if (cache->cached != BTRFS_CACHE_NO) {
515 spin_unlock(&cache->lock);
519 WARN_ON(cache->caching_ctl);
520 cache->caching_ctl = caching_ctl;
521 cache->cached = BTRFS_CACHE_FAST;
522 spin_unlock(&cache->lock);
525 * We can't do the read from on-disk cache during a commit since we need
526 * to have the normal tree locking. Also if we are currently trying to
527 * allocate blocks for the tree root we can't do the fast caching since
528 * we likely hold important locks.
530 if (trans && (!trans->transaction->in_commit) &&
531 (root && root != root->fs_info->tree_root) &&
532 btrfs_test_opt(root, SPACE_CACHE)) {
533 ret = load_free_space_cache(fs_info, cache);
535 spin_lock(&cache->lock);
537 cache->caching_ctl = NULL;
538 cache->cached = BTRFS_CACHE_FINISHED;
539 cache->last_byte_to_unpin = (u64)-1;
541 if (load_cache_only) {
542 cache->caching_ctl = NULL;
543 cache->cached = BTRFS_CACHE_NO;
545 cache->cached = BTRFS_CACHE_STARTED;
548 spin_unlock(&cache->lock);
549 wake_up(&caching_ctl->wait);
551 put_caching_control(caching_ctl);
552 free_excluded_extents(fs_info->extent_root, cache);
557 * We are not going to do the fast caching, set cached to the
558 * appropriate value and wakeup any waiters.
560 spin_lock(&cache->lock);
561 if (load_cache_only) {
562 cache->caching_ctl = NULL;
563 cache->cached = BTRFS_CACHE_NO;
565 cache->cached = BTRFS_CACHE_STARTED;
567 spin_unlock(&cache->lock);
568 wake_up(&caching_ctl->wait);
571 if (load_cache_only) {
572 put_caching_control(caching_ctl);
576 down_write(&fs_info->extent_commit_sem);
577 atomic_inc(&caching_ctl->count);
578 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
579 up_write(&fs_info->extent_commit_sem);
581 btrfs_get_block_group(cache);
583 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
589 * return the block group that starts at or after bytenr
591 static struct btrfs_block_group_cache *
592 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
594 struct btrfs_block_group_cache *cache;
596 cache = block_group_cache_tree_search(info, bytenr, 0);
602 * return the block group that contains the given bytenr
604 struct btrfs_block_group_cache *btrfs_lookup_block_group(
605 struct btrfs_fs_info *info,
608 struct btrfs_block_group_cache *cache;
610 cache = block_group_cache_tree_search(info, bytenr, 1);
615 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
618 struct list_head *head = &info->space_info;
619 struct btrfs_space_info *found;
621 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
624 list_for_each_entry_rcu(found, head, list) {
625 if (found->flags & flags) {
635 * after adding space to the filesystem, we need to clear the full flags
636 * on all the space infos.
638 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
640 struct list_head *head = &info->space_info;
641 struct btrfs_space_info *found;
644 list_for_each_entry_rcu(found, head, list)
649 static u64 div_factor(u64 num, int factor)
658 static u64 div_factor_fine(u64 num, int factor)
667 u64 btrfs_find_block_group(struct btrfs_root *root,
668 u64 search_start, u64 search_hint, int owner)
670 struct btrfs_block_group_cache *cache;
672 u64 last = max(search_hint, search_start);
679 cache = btrfs_lookup_first_block_group(root->fs_info, last);
683 spin_lock(&cache->lock);
684 last = cache->key.objectid + cache->key.offset;
685 used = btrfs_block_group_used(&cache->item);
687 if ((full_search || !cache->ro) &&
688 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
689 if (used + cache->pinned + cache->reserved <
690 div_factor(cache->key.offset, factor)) {
691 group_start = cache->key.objectid;
692 spin_unlock(&cache->lock);
693 btrfs_put_block_group(cache);
697 spin_unlock(&cache->lock);
698 btrfs_put_block_group(cache);
706 if (!full_search && factor < 10) {
716 /* simple helper to search for an existing extent at a given offset */
717 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
720 struct btrfs_key key;
721 struct btrfs_path *path;
723 path = btrfs_alloc_path();
727 key.objectid = start;
729 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
730 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
732 btrfs_free_path(path);
737 * helper function to lookup reference count and flags of extent.
739 * the head node for delayed ref is used to store the sum of all the
740 * reference count modifications queued up in the rbtree. the head
741 * node may also store the extent flags to set. This way you can check
742 * to see what the reference count and extent flags would be if all of
743 * the delayed refs are not processed.
745 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
746 struct btrfs_root *root, u64 bytenr,
747 u64 num_bytes, u64 *refs, u64 *flags)
749 struct btrfs_delayed_ref_head *head;
750 struct btrfs_delayed_ref_root *delayed_refs;
751 struct btrfs_path *path;
752 struct btrfs_extent_item *ei;
753 struct extent_buffer *leaf;
754 struct btrfs_key key;
760 path = btrfs_alloc_path();
764 key.objectid = bytenr;
765 key.type = BTRFS_EXTENT_ITEM_KEY;
766 key.offset = num_bytes;
768 path->skip_locking = 1;
769 path->search_commit_root = 1;
772 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
778 leaf = path->nodes[0];
779 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
780 if (item_size >= sizeof(*ei)) {
781 ei = btrfs_item_ptr(leaf, path->slots[0],
782 struct btrfs_extent_item);
783 num_refs = btrfs_extent_refs(leaf, ei);
784 extent_flags = btrfs_extent_flags(leaf, ei);
786 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
787 struct btrfs_extent_item_v0 *ei0;
788 BUG_ON(item_size != sizeof(*ei0));
789 ei0 = btrfs_item_ptr(leaf, path->slots[0],
790 struct btrfs_extent_item_v0);
791 num_refs = btrfs_extent_refs_v0(leaf, ei0);
792 /* FIXME: this isn't correct for data */
793 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
798 BUG_ON(num_refs == 0);
808 delayed_refs = &trans->transaction->delayed_refs;
809 spin_lock(&delayed_refs->lock);
810 head = btrfs_find_delayed_ref_head(trans, bytenr);
812 if (!mutex_trylock(&head->mutex)) {
813 atomic_inc(&head->node.refs);
814 spin_unlock(&delayed_refs->lock);
816 btrfs_release_path(path);
819 * Mutex was contended, block until it's released and try
822 mutex_lock(&head->mutex);
823 mutex_unlock(&head->mutex);
824 btrfs_put_delayed_ref(&head->node);
827 if (head->extent_op && head->extent_op->update_flags)
828 extent_flags |= head->extent_op->flags_to_set;
830 BUG_ON(num_refs == 0);
832 num_refs += head->node.ref_mod;
833 mutex_unlock(&head->mutex);
835 spin_unlock(&delayed_refs->lock);
837 WARN_ON(num_refs == 0);
841 *flags = extent_flags;
843 btrfs_free_path(path);
848 * Back reference rules. Back refs have three main goals:
850 * 1) differentiate between all holders of references to an extent so that
851 * when a reference is dropped we can make sure it was a valid reference
852 * before freeing the extent.
854 * 2) Provide enough information to quickly find the holders of an extent
855 * if we notice a given block is corrupted or bad.
857 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
858 * maintenance. This is actually the same as #2, but with a slightly
859 * different use case.
861 * There are two kinds of back refs. The implicit back refs is optimized
862 * for pointers in non-shared tree blocks. For a given pointer in a block,
863 * back refs of this kind provide information about the block's owner tree
864 * and the pointer's key. These information allow us to find the block by
865 * b-tree searching. The full back refs is for pointers in tree blocks not
866 * referenced by their owner trees. The location of tree block is recorded
867 * in the back refs. Actually the full back refs is generic, and can be
868 * used in all cases the implicit back refs is used. The major shortcoming
869 * of the full back refs is its overhead. Every time a tree block gets
870 * COWed, we have to update back refs entry for all pointers in it.
872 * For a newly allocated tree block, we use implicit back refs for
873 * pointers in it. This means most tree related operations only involve
874 * implicit back refs. For a tree block created in old transaction, the
875 * only way to drop a reference to it is COW it. So we can detect the
876 * event that tree block loses its owner tree's reference and do the
877 * back refs conversion.
879 * When a tree block is COW'd through a tree, there are four cases:
881 * The reference count of the block is one and the tree is the block's
882 * owner tree. Nothing to do in this case.
884 * The reference count of the block is one and the tree is not the
885 * block's owner tree. In this case, full back refs is used for pointers
886 * in the block. Remove these full back refs, add implicit back refs for
887 * every pointers in the new block.
889 * The reference count of the block is greater than one and the tree is
890 * the block's owner tree. In this case, implicit back refs is used for
891 * pointers in the block. Add full back refs for every pointers in the
892 * block, increase lower level extents' reference counts. The original
893 * implicit back refs are entailed to the new block.
895 * The reference count of the block is greater than one and the tree is
896 * not the block's owner tree. Add implicit back refs for every pointer in
897 * the new block, increase lower level extents' reference count.
899 * Back Reference Key composing:
901 * The key objectid corresponds to the first byte in the extent,
902 * The key type is used to differentiate between types of back refs.
903 * There are different meanings of the key offset for different types
906 * File extents can be referenced by:
908 * - multiple snapshots, subvolumes, or different generations in one subvol
909 * - different files inside a single subvolume
910 * - different offsets inside a file (bookend extents in file.c)
912 * The extent ref structure for the implicit back refs has fields for:
914 * - Objectid of the subvolume root
915 * - objectid of the file holding the reference
916 * - original offset in the file
917 * - how many bookend extents
919 * The key offset for the implicit back refs is hash of the first
922 * The extent ref structure for the full back refs has field for:
924 * - number of pointers in the tree leaf
926 * The key offset for the implicit back refs is the first byte of
929 * When a file extent is allocated, The implicit back refs is used.
930 * the fields are filled in:
932 * (root_key.objectid, inode objectid, offset in file, 1)
934 * When a file extent is removed file truncation, we find the
935 * corresponding implicit back refs and check the following fields:
937 * (btrfs_header_owner(leaf), inode objectid, offset in file)
939 * Btree extents can be referenced by:
941 * - Different subvolumes
943 * Both the implicit back refs and the full back refs for tree blocks
944 * only consist of key. The key offset for the implicit back refs is
945 * objectid of block's owner tree. The key offset for the full back refs
946 * is the first byte of parent block.
948 * When implicit back refs is used, information about the lowest key and
949 * level of the tree block are required. These information are stored in
950 * tree block info structure.
953 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
954 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
955 struct btrfs_root *root,
956 struct btrfs_path *path,
957 u64 owner, u32 extra_size)
959 struct btrfs_extent_item *item;
960 struct btrfs_extent_item_v0 *ei0;
961 struct btrfs_extent_ref_v0 *ref0;
962 struct btrfs_tree_block_info *bi;
963 struct extent_buffer *leaf;
964 struct btrfs_key key;
965 struct btrfs_key found_key;
966 u32 new_size = sizeof(*item);
970 leaf = path->nodes[0];
971 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
973 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
974 ei0 = btrfs_item_ptr(leaf, path->slots[0],
975 struct btrfs_extent_item_v0);
976 refs = btrfs_extent_refs_v0(leaf, ei0);
978 if (owner == (u64)-1) {
980 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
981 ret = btrfs_next_leaf(root, path);
985 leaf = path->nodes[0];
987 btrfs_item_key_to_cpu(leaf, &found_key,
989 BUG_ON(key.objectid != found_key.objectid);
990 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
994 ref0 = btrfs_item_ptr(leaf, path->slots[0],
995 struct btrfs_extent_ref_v0);
996 owner = btrfs_ref_objectid_v0(leaf, ref0);
1000 btrfs_release_path(path);
1002 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1003 new_size += sizeof(*bi);
1005 new_size -= sizeof(*ei0);
1006 ret = btrfs_search_slot(trans, root, &key, path,
1007 new_size + extra_size, 1);
1012 ret = btrfs_extend_item(trans, root, path, new_size);
1014 leaf = path->nodes[0];
1015 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1016 btrfs_set_extent_refs(leaf, item, refs);
1017 /* FIXME: get real generation */
1018 btrfs_set_extent_generation(leaf, item, 0);
1019 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1020 btrfs_set_extent_flags(leaf, item,
1021 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1022 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1023 bi = (struct btrfs_tree_block_info *)(item + 1);
1024 /* FIXME: get first key of the block */
1025 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1026 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1028 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1030 btrfs_mark_buffer_dirty(leaf);
1035 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1037 u32 high_crc = ~(u32)0;
1038 u32 low_crc = ~(u32)0;
1041 lenum = cpu_to_le64(root_objectid);
1042 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1043 lenum = cpu_to_le64(owner);
1044 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1045 lenum = cpu_to_le64(offset);
1046 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1048 return ((u64)high_crc << 31) ^ (u64)low_crc;
1051 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1052 struct btrfs_extent_data_ref *ref)
1054 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1055 btrfs_extent_data_ref_objectid(leaf, ref),
1056 btrfs_extent_data_ref_offset(leaf, ref));
1059 static int match_extent_data_ref(struct extent_buffer *leaf,
1060 struct btrfs_extent_data_ref *ref,
1061 u64 root_objectid, u64 owner, u64 offset)
1063 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1064 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1065 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1070 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1071 struct btrfs_root *root,
1072 struct btrfs_path *path,
1073 u64 bytenr, u64 parent,
1075 u64 owner, u64 offset)
1077 struct btrfs_key key;
1078 struct btrfs_extent_data_ref *ref;
1079 struct extent_buffer *leaf;
1085 key.objectid = bytenr;
1087 key.type = BTRFS_SHARED_DATA_REF_KEY;
1088 key.offset = parent;
1090 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1091 key.offset = hash_extent_data_ref(root_objectid,
1096 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106 key.type = BTRFS_EXTENT_REF_V0_KEY;
1107 btrfs_release_path(path);
1108 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1119 leaf = path->nodes[0];
1120 nritems = btrfs_header_nritems(leaf);
1122 if (path->slots[0] >= nritems) {
1123 ret = btrfs_next_leaf(root, path);
1129 leaf = path->nodes[0];
1130 nritems = btrfs_header_nritems(leaf);
1134 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1135 if (key.objectid != bytenr ||
1136 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1139 ref = btrfs_item_ptr(leaf, path->slots[0],
1140 struct btrfs_extent_data_ref);
1142 if (match_extent_data_ref(leaf, ref, root_objectid,
1145 btrfs_release_path(path);
1157 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1158 struct btrfs_root *root,
1159 struct btrfs_path *path,
1160 u64 bytenr, u64 parent,
1161 u64 root_objectid, u64 owner,
1162 u64 offset, int refs_to_add)
1164 struct btrfs_key key;
1165 struct extent_buffer *leaf;
1170 key.objectid = bytenr;
1172 key.type = BTRFS_SHARED_DATA_REF_KEY;
1173 key.offset = parent;
1174 size = sizeof(struct btrfs_shared_data_ref);
1176 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1177 key.offset = hash_extent_data_ref(root_objectid,
1179 size = sizeof(struct btrfs_extent_data_ref);
1182 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1183 if (ret && ret != -EEXIST)
1186 leaf = path->nodes[0];
1188 struct btrfs_shared_data_ref *ref;
1189 ref = btrfs_item_ptr(leaf, path->slots[0],
1190 struct btrfs_shared_data_ref);
1192 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1194 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1195 num_refs += refs_to_add;
1196 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1199 struct btrfs_extent_data_ref *ref;
1200 while (ret == -EEXIST) {
1201 ref = btrfs_item_ptr(leaf, path->slots[0],
1202 struct btrfs_extent_data_ref);
1203 if (match_extent_data_ref(leaf, ref, root_objectid,
1206 btrfs_release_path(path);
1208 ret = btrfs_insert_empty_item(trans, root, path, &key,
1210 if (ret && ret != -EEXIST)
1213 leaf = path->nodes[0];
1215 ref = btrfs_item_ptr(leaf, path->slots[0],
1216 struct btrfs_extent_data_ref);
1218 btrfs_set_extent_data_ref_root(leaf, ref,
1220 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1221 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1222 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1224 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1225 num_refs += refs_to_add;
1226 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1229 btrfs_mark_buffer_dirty(leaf);
1232 btrfs_release_path(path);
1236 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1237 struct btrfs_root *root,
1238 struct btrfs_path *path,
1241 struct btrfs_key key;
1242 struct btrfs_extent_data_ref *ref1 = NULL;
1243 struct btrfs_shared_data_ref *ref2 = NULL;
1244 struct extent_buffer *leaf;
1248 leaf = path->nodes[0];
1249 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1251 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1252 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1253 struct btrfs_extent_data_ref);
1254 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1255 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1256 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1257 struct btrfs_shared_data_ref);
1258 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1259 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1260 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1261 struct btrfs_extent_ref_v0 *ref0;
1262 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1263 struct btrfs_extent_ref_v0);
1264 num_refs = btrfs_ref_count_v0(leaf, ref0);
1270 BUG_ON(num_refs < refs_to_drop);
1271 num_refs -= refs_to_drop;
1273 if (num_refs == 0) {
1274 ret = btrfs_del_item(trans, root, path);
1276 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1277 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1278 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1279 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1282 struct btrfs_extent_ref_v0 *ref0;
1283 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1284 struct btrfs_extent_ref_v0);
1285 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1288 btrfs_mark_buffer_dirty(leaf);
1293 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1294 struct btrfs_path *path,
1295 struct btrfs_extent_inline_ref *iref)
1297 struct btrfs_key key;
1298 struct extent_buffer *leaf;
1299 struct btrfs_extent_data_ref *ref1;
1300 struct btrfs_shared_data_ref *ref2;
1303 leaf = path->nodes[0];
1304 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1306 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1307 BTRFS_EXTENT_DATA_REF_KEY) {
1308 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1309 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1311 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1312 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1314 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1315 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1316 struct btrfs_extent_data_ref);
1317 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1318 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1319 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_shared_data_ref);
1321 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1322 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1323 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1324 struct btrfs_extent_ref_v0 *ref0;
1325 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1326 struct btrfs_extent_ref_v0);
1327 num_refs = btrfs_ref_count_v0(leaf, ref0);
1335 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1336 struct btrfs_root *root,
1337 struct btrfs_path *path,
1338 u64 bytenr, u64 parent,
1341 struct btrfs_key key;
1344 key.objectid = bytenr;
1346 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1347 key.offset = parent;
1349 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1350 key.offset = root_objectid;
1353 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357 if (ret == -ENOENT && parent) {
1358 btrfs_release_path(path);
1359 key.type = BTRFS_EXTENT_REF_V0_KEY;
1360 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1368 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1369 struct btrfs_root *root,
1370 struct btrfs_path *path,
1371 u64 bytenr, u64 parent,
1374 struct btrfs_key key;
1377 key.objectid = bytenr;
1379 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1380 key.offset = parent;
1382 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1383 key.offset = root_objectid;
1386 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1387 btrfs_release_path(path);
1391 static inline int extent_ref_type(u64 parent, u64 owner)
1394 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1396 type = BTRFS_SHARED_BLOCK_REF_KEY;
1398 type = BTRFS_TREE_BLOCK_REF_KEY;
1401 type = BTRFS_SHARED_DATA_REF_KEY;
1403 type = BTRFS_EXTENT_DATA_REF_KEY;
1408 static int find_next_key(struct btrfs_path *path, int level,
1409 struct btrfs_key *key)
1412 for (; level < BTRFS_MAX_LEVEL; level++) {
1413 if (!path->nodes[level])
1415 if (path->slots[level] + 1 >=
1416 btrfs_header_nritems(path->nodes[level]))
1419 btrfs_item_key_to_cpu(path->nodes[level], key,
1420 path->slots[level] + 1);
1422 btrfs_node_key_to_cpu(path->nodes[level], key,
1423 path->slots[level] + 1);
1430 * look for inline back ref. if back ref is found, *ref_ret is set
1431 * to the address of inline back ref, and 0 is returned.
1433 * if back ref isn't found, *ref_ret is set to the address where it
1434 * should be inserted, and -ENOENT is returned.
1436 * if insert is true and there are too many inline back refs, the path
1437 * points to the extent item, and -EAGAIN is returned.
1439 * NOTE: inline back refs are ordered in the same way that back ref
1440 * items in the tree are ordered.
1442 static noinline_for_stack
1443 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1444 struct btrfs_root *root,
1445 struct btrfs_path *path,
1446 struct btrfs_extent_inline_ref **ref_ret,
1447 u64 bytenr, u64 num_bytes,
1448 u64 parent, u64 root_objectid,
1449 u64 owner, u64 offset, int insert)
1451 struct btrfs_key key;
1452 struct extent_buffer *leaf;
1453 struct btrfs_extent_item *ei;
1454 struct btrfs_extent_inline_ref *iref;
1465 key.objectid = bytenr;
1466 key.type = BTRFS_EXTENT_ITEM_KEY;
1467 key.offset = num_bytes;
1469 want = extent_ref_type(parent, owner);
1471 extra_size = btrfs_extent_inline_ref_size(want);
1472 path->keep_locks = 1;
1475 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1482 leaf = path->nodes[0];
1483 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1484 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1485 if (item_size < sizeof(*ei)) {
1490 ret = convert_extent_item_v0(trans, root, path, owner,
1496 leaf = path->nodes[0];
1497 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1500 BUG_ON(item_size < sizeof(*ei));
1502 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1503 flags = btrfs_extent_flags(leaf, ei);
1505 ptr = (unsigned long)(ei + 1);
1506 end = (unsigned long)ei + item_size;
1508 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1509 ptr += sizeof(struct btrfs_tree_block_info);
1512 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1521 iref = (struct btrfs_extent_inline_ref *)ptr;
1522 type = btrfs_extent_inline_ref_type(leaf, iref);
1526 ptr += btrfs_extent_inline_ref_size(type);
1530 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1531 struct btrfs_extent_data_ref *dref;
1532 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1533 if (match_extent_data_ref(leaf, dref, root_objectid,
1538 if (hash_extent_data_ref_item(leaf, dref) <
1539 hash_extent_data_ref(root_objectid, owner, offset))
1543 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1545 if (parent == ref_offset) {
1549 if (ref_offset < parent)
1552 if (root_objectid == ref_offset) {
1556 if (ref_offset < root_objectid)
1560 ptr += btrfs_extent_inline_ref_size(type);
1562 if (err == -ENOENT && insert) {
1563 if (item_size + extra_size >=
1564 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1569 * To add new inline back ref, we have to make sure
1570 * there is no corresponding back ref item.
1571 * For simplicity, we just do not add new inline back
1572 * ref if there is any kind of item for this block
1574 if (find_next_key(path, 0, &key) == 0 &&
1575 key.objectid == bytenr &&
1576 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1581 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1584 path->keep_locks = 0;
1585 btrfs_unlock_up_safe(path, 1);
1591 * helper to add new inline back ref
1593 static noinline_for_stack
1594 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1595 struct btrfs_root *root,
1596 struct btrfs_path *path,
1597 struct btrfs_extent_inline_ref *iref,
1598 u64 parent, u64 root_objectid,
1599 u64 owner, u64 offset, int refs_to_add,
1600 struct btrfs_delayed_extent_op *extent_op)
1602 struct extent_buffer *leaf;
1603 struct btrfs_extent_item *ei;
1606 unsigned long item_offset;
1612 leaf = path->nodes[0];
1613 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1614 item_offset = (unsigned long)iref - (unsigned long)ei;
1616 type = extent_ref_type(parent, owner);
1617 size = btrfs_extent_inline_ref_size(type);
1619 ret = btrfs_extend_item(trans, root, path, size);
1621 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1622 refs = btrfs_extent_refs(leaf, ei);
1623 refs += refs_to_add;
1624 btrfs_set_extent_refs(leaf, ei, refs);
1626 __run_delayed_extent_op(extent_op, leaf, ei);
1628 ptr = (unsigned long)ei + item_offset;
1629 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1630 if (ptr < end - size)
1631 memmove_extent_buffer(leaf, ptr + size, ptr,
1634 iref = (struct btrfs_extent_inline_ref *)ptr;
1635 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1636 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1637 struct btrfs_extent_data_ref *dref;
1638 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1639 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1640 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1641 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1642 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1643 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1644 struct btrfs_shared_data_ref *sref;
1645 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1646 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1647 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1648 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1649 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1651 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1653 btrfs_mark_buffer_dirty(leaf);
1657 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1658 struct btrfs_root *root,
1659 struct btrfs_path *path,
1660 struct btrfs_extent_inline_ref **ref_ret,
1661 u64 bytenr, u64 num_bytes, u64 parent,
1662 u64 root_objectid, u64 owner, u64 offset)
1666 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1667 bytenr, num_bytes, parent,
1668 root_objectid, owner, offset, 0);
1672 btrfs_release_path(path);
1675 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1676 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1679 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1680 root_objectid, owner, offset);
1686 * helper to update/remove inline back ref
1688 static noinline_for_stack
1689 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1690 struct btrfs_root *root,
1691 struct btrfs_path *path,
1692 struct btrfs_extent_inline_ref *iref,
1694 struct btrfs_delayed_extent_op *extent_op)
1696 struct extent_buffer *leaf;
1697 struct btrfs_extent_item *ei;
1698 struct btrfs_extent_data_ref *dref = NULL;
1699 struct btrfs_shared_data_ref *sref = NULL;
1708 leaf = path->nodes[0];
1709 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1710 refs = btrfs_extent_refs(leaf, ei);
1711 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1712 refs += refs_to_mod;
1713 btrfs_set_extent_refs(leaf, ei, refs);
1715 __run_delayed_extent_op(extent_op, leaf, ei);
1717 type = btrfs_extent_inline_ref_type(leaf, iref);
1719 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1720 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1721 refs = btrfs_extent_data_ref_count(leaf, dref);
1722 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1723 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1724 refs = btrfs_shared_data_ref_count(leaf, sref);
1727 BUG_ON(refs_to_mod != -1);
1730 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1731 refs += refs_to_mod;
1734 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1735 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1737 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1739 size = btrfs_extent_inline_ref_size(type);
1740 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1741 ptr = (unsigned long)iref;
1742 end = (unsigned long)ei + item_size;
1743 if (ptr + size < end)
1744 memmove_extent_buffer(leaf, ptr, ptr + size,
1747 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1749 btrfs_mark_buffer_dirty(leaf);
1753 static noinline_for_stack
1754 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1755 struct btrfs_root *root,
1756 struct btrfs_path *path,
1757 u64 bytenr, u64 num_bytes, u64 parent,
1758 u64 root_objectid, u64 owner,
1759 u64 offset, int refs_to_add,
1760 struct btrfs_delayed_extent_op *extent_op)
1762 struct btrfs_extent_inline_ref *iref;
1765 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1766 bytenr, num_bytes, parent,
1767 root_objectid, owner, offset, 1);
1769 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1770 ret = update_inline_extent_backref(trans, root, path, iref,
1771 refs_to_add, extent_op);
1772 } else if (ret == -ENOENT) {
1773 ret = setup_inline_extent_backref(trans, root, path, iref,
1774 parent, root_objectid,
1775 owner, offset, refs_to_add,
1781 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1782 struct btrfs_root *root,
1783 struct btrfs_path *path,
1784 u64 bytenr, u64 parent, u64 root_objectid,
1785 u64 owner, u64 offset, int refs_to_add)
1788 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1789 BUG_ON(refs_to_add != 1);
1790 ret = insert_tree_block_ref(trans, root, path, bytenr,
1791 parent, root_objectid);
1793 ret = insert_extent_data_ref(trans, root, path, bytenr,
1794 parent, root_objectid,
1795 owner, offset, refs_to_add);
1800 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1801 struct btrfs_root *root,
1802 struct btrfs_path *path,
1803 struct btrfs_extent_inline_ref *iref,
1804 int refs_to_drop, int is_data)
1808 BUG_ON(!is_data && refs_to_drop != 1);
1810 ret = update_inline_extent_backref(trans, root, path, iref,
1811 -refs_to_drop, NULL);
1812 } else if (is_data) {
1813 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1815 ret = btrfs_del_item(trans, root, path);
1820 static int btrfs_issue_discard(struct block_device *bdev,
1823 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1826 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1827 u64 num_bytes, u64 *actual_bytes)
1830 u64 discarded_bytes = 0;
1831 struct btrfs_bio *bbio = NULL;
1834 /* Tell the block device(s) that the sectors can be discarded */
1835 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1836 bytenr, &num_bytes, &bbio, 0);
1838 struct btrfs_bio_stripe *stripe = bbio->stripes;
1842 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1843 if (!stripe->dev->can_discard)
1846 ret = btrfs_issue_discard(stripe->dev->bdev,
1850 discarded_bytes += stripe->length;
1851 else if (ret != -EOPNOTSUPP)
1855 * Just in case we get back EOPNOTSUPP for some reason,
1856 * just ignore the return value so we don't screw up
1857 * people calling discard_extent.
1865 *actual_bytes = discarded_bytes;
1871 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1872 struct btrfs_root *root,
1873 u64 bytenr, u64 num_bytes, u64 parent,
1874 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1877 struct btrfs_fs_info *fs_info = root->fs_info;
1879 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1880 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1882 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1883 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1885 parent, root_objectid, (int)owner,
1886 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1888 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1890 parent, root_objectid, owner, offset,
1891 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1896 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1897 struct btrfs_root *root,
1898 u64 bytenr, u64 num_bytes,
1899 u64 parent, u64 root_objectid,
1900 u64 owner, u64 offset, int refs_to_add,
1901 struct btrfs_delayed_extent_op *extent_op)
1903 struct btrfs_path *path;
1904 struct extent_buffer *leaf;
1905 struct btrfs_extent_item *item;
1910 path = btrfs_alloc_path();
1915 path->leave_spinning = 1;
1916 /* this will setup the path even if it fails to insert the back ref */
1917 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1918 path, bytenr, num_bytes, parent,
1919 root_objectid, owner, offset,
1920 refs_to_add, extent_op);
1924 if (ret != -EAGAIN) {
1929 leaf = path->nodes[0];
1930 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1931 refs = btrfs_extent_refs(leaf, item);
1932 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1934 __run_delayed_extent_op(extent_op, leaf, item);
1936 btrfs_mark_buffer_dirty(leaf);
1937 btrfs_release_path(path);
1940 path->leave_spinning = 1;
1942 /* now insert the actual backref */
1943 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1944 path, bytenr, parent, root_objectid,
1945 owner, offset, refs_to_add);
1948 btrfs_free_path(path);
1952 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1953 struct btrfs_root *root,
1954 struct btrfs_delayed_ref_node *node,
1955 struct btrfs_delayed_extent_op *extent_op,
1956 int insert_reserved)
1959 struct btrfs_delayed_data_ref *ref;
1960 struct btrfs_key ins;
1965 ins.objectid = node->bytenr;
1966 ins.offset = node->num_bytes;
1967 ins.type = BTRFS_EXTENT_ITEM_KEY;
1969 ref = btrfs_delayed_node_to_data_ref(node);
1970 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1971 parent = ref->parent;
1973 ref_root = ref->root;
1975 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1977 BUG_ON(extent_op->update_key);
1978 flags |= extent_op->flags_to_set;
1980 ret = alloc_reserved_file_extent(trans, root,
1981 parent, ref_root, flags,
1982 ref->objectid, ref->offset,
1983 &ins, node->ref_mod);
1984 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1985 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1986 node->num_bytes, parent,
1987 ref_root, ref->objectid,
1988 ref->offset, node->ref_mod,
1990 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1991 ret = __btrfs_free_extent(trans, root, node->bytenr,
1992 node->num_bytes, parent,
1993 ref_root, ref->objectid,
1994 ref->offset, node->ref_mod,
2002 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2003 struct extent_buffer *leaf,
2004 struct btrfs_extent_item *ei)
2006 u64 flags = btrfs_extent_flags(leaf, ei);
2007 if (extent_op->update_flags) {
2008 flags |= extent_op->flags_to_set;
2009 btrfs_set_extent_flags(leaf, ei, flags);
2012 if (extent_op->update_key) {
2013 struct btrfs_tree_block_info *bi;
2014 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2015 bi = (struct btrfs_tree_block_info *)(ei + 1);
2016 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2020 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2021 struct btrfs_root *root,
2022 struct btrfs_delayed_ref_node *node,
2023 struct btrfs_delayed_extent_op *extent_op)
2025 struct btrfs_key key;
2026 struct btrfs_path *path;
2027 struct btrfs_extent_item *ei;
2028 struct extent_buffer *leaf;
2033 path = btrfs_alloc_path();
2037 key.objectid = node->bytenr;
2038 key.type = BTRFS_EXTENT_ITEM_KEY;
2039 key.offset = node->num_bytes;
2042 path->leave_spinning = 1;
2043 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2054 leaf = path->nodes[0];
2055 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2056 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2057 if (item_size < sizeof(*ei)) {
2058 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2064 leaf = path->nodes[0];
2065 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2068 BUG_ON(item_size < sizeof(*ei));
2069 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2070 __run_delayed_extent_op(extent_op, leaf, ei);
2072 btrfs_mark_buffer_dirty(leaf);
2074 btrfs_free_path(path);
2078 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2079 struct btrfs_root *root,
2080 struct btrfs_delayed_ref_node *node,
2081 struct btrfs_delayed_extent_op *extent_op,
2082 int insert_reserved)
2085 struct btrfs_delayed_tree_ref *ref;
2086 struct btrfs_key ins;
2090 ins.objectid = node->bytenr;
2091 ins.offset = node->num_bytes;
2092 ins.type = BTRFS_EXTENT_ITEM_KEY;
2094 ref = btrfs_delayed_node_to_tree_ref(node);
2095 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2096 parent = ref->parent;
2098 ref_root = ref->root;
2100 BUG_ON(node->ref_mod != 1);
2101 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2102 BUG_ON(!extent_op || !extent_op->update_flags ||
2103 !extent_op->update_key);
2104 ret = alloc_reserved_tree_block(trans, root,
2106 extent_op->flags_to_set,
2109 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2110 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2111 node->num_bytes, parent, ref_root,
2112 ref->level, 0, 1, extent_op);
2113 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2114 ret = __btrfs_free_extent(trans, root, node->bytenr,
2115 node->num_bytes, parent, ref_root,
2116 ref->level, 0, 1, extent_op);
2123 /* helper function to actually process a single delayed ref entry */
2124 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2125 struct btrfs_root *root,
2126 struct btrfs_delayed_ref_node *node,
2127 struct btrfs_delayed_extent_op *extent_op,
2128 int insert_reserved)
2131 if (btrfs_delayed_ref_is_head(node)) {
2132 struct btrfs_delayed_ref_head *head;
2134 * we've hit the end of the chain and we were supposed
2135 * to insert this extent into the tree. But, it got
2136 * deleted before we ever needed to insert it, so all
2137 * we have to do is clean up the accounting
2140 head = btrfs_delayed_node_to_head(node);
2141 if (insert_reserved) {
2142 btrfs_pin_extent(root, node->bytenr,
2143 node->num_bytes, 1);
2144 if (head->is_data) {
2145 ret = btrfs_del_csums(trans, root,
2151 mutex_unlock(&head->mutex);
2155 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2156 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2157 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2159 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2160 node->type == BTRFS_SHARED_DATA_REF_KEY)
2161 ret = run_delayed_data_ref(trans, root, node, extent_op,
2168 static noinline struct btrfs_delayed_ref_node *
2169 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2171 struct rb_node *node;
2172 struct btrfs_delayed_ref_node *ref;
2173 int action = BTRFS_ADD_DELAYED_REF;
2176 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2177 * this prevents ref count from going down to zero when
2178 * there still are pending delayed ref.
2180 node = rb_prev(&head->node.rb_node);
2184 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2186 if (ref->bytenr != head->node.bytenr)
2188 if (ref->action == action)
2190 node = rb_prev(node);
2192 if (action == BTRFS_ADD_DELAYED_REF) {
2193 action = BTRFS_DROP_DELAYED_REF;
2199 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2200 struct btrfs_root *root,
2201 struct list_head *cluster)
2203 struct btrfs_delayed_ref_root *delayed_refs;
2204 struct btrfs_delayed_ref_node *ref;
2205 struct btrfs_delayed_ref_head *locked_ref = NULL;
2206 struct btrfs_delayed_extent_op *extent_op;
2209 int must_insert_reserved = 0;
2211 delayed_refs = &trans->transaction->delayed_refs;
2214 /* pick a new head ref from the cluster list */
2215 if (list_empty(cluster))
2218 locked_ref = list_entry(cluster->next,
2219 struct btrfs_delayed_ref_head, cluster);
2221 /* grab the lock that says we are going to process
2222 * all the refs for this head */
2223 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2226 * we may have dropped the spin lock to get the head
2227 * mutex lock, and that might have given someone else
2228 * time to free the head. If that's true, it has been
2229 * removed from our list and we can move on.
2231 if (ret == -EAGAIN) {
2239 * locked_ref is the head node, so we have to go one
2240 * node back for any delayed ref updates
2242 ref = select_delayed_ref(locked_ref);
2244 if (ref && ref->seq &&
2245 btrfs_check_delayed_seq(delayed_refs, ref->seq)) {
2247 * there are still refs with lower seq numbers in the
2248 * process of being added. Don't run this ref yet.
2250 list_del_init(&locked_ref->cluster);
2251 mutex_unlock(&locked_ref->mutex);
2253 delayed_refs->num_heads_ready++;
2254 spin_unlock(&delayed_refs->lock);
2256 spin_lock(&delayed_refs->lock);
2261 * record the must insert reserved flag before we
2262 * drop the spin lock.
2264 must_insert_reserved = locked_ref->must_insert_reserved;
2265 locked_ref->must_insert_reserved = 0;
2267 extent_op = locked_ref->extent_op;
2268 locked_ref->extent_op = NULL;
2271 /* All delayed refs have been processed, Go ahead
2272 * and send the head node to run_one_delayed_ref,
2273 * so that any accounting fixes can happen
2275 ref = &locked_ref->node;
2277 if (extent_op && must_insert_reserved) {
2283 spin_unlock(&delayed_refs->lock);
2285 ret = run_delayed_extent_op(trans, root,
2293 list_del_init(&locked_ref->cluster);
2298 rb_erase(&ref->rb_node, &delayed_refs->root);
2299 delayed_refs->num_entries--;
2301 * we modified num_entries, but as we're currently running
2302 * delayed refs, skip
2303 * wake_up(&delayed_refs->seq_wait);
2306 spin_unlock(&delayed_refs->lock);
2308 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2309 must_insert_reserved);
2312 btrfs_put_delayed_ref(ref);
2316 do_chunk_alloc(trans, root->fs_info->extent_root,
2318 btrfs_get_alloc_profile(root, 0),
2319 CHUNK_ALLOC_NO_FORCE);
2321 spin_lock(&delayed_refs->lock);
2327 static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
2328 unsigned long num_refs)
2330 struct list_head *first_seq = delayed_refs->seq_head.next;
2332 spin_unlock(&delayed_refs->lock);
2333 pr_debug("waiting for more refs (num %ld, first %p)\n",
2334 num_refs, first_seq);
2335 wait_event(delayed_refs->seq_wait,
2336 num_refs != delayed_refs->num_entries ||
2337 delayed_refs->seq_head.next != first_seq);
2338 pr_debug("done waiting for more refs (num %ld, first %p)\n",
2339 delayed_refs->num_entries, delayed_refs->seq_head.next);
2340 spin_lock(&delayed_refs->lock);
2344 * this starts processing the delayed reference count updates and
2345 * extent insertions we have queued up so far. count can be
2346 * 0, which means to process everything in the tree at the start
2347 * of the run (but not newly added entries), or it can be some target
2348 * number you'd like to process.
2350 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2351 struct btrfs_root *root, unsigned long count)
2353 struct rb_node *node;
2354 struct btrfs_delayed_ref_root *delayed_refs;
2355 struct btrfs_delayed_ref_node *ref;
2356 struct list_head cluster;
2359 int run_all = count == (unsigned long)-1;
2361 unsigned long num_refs = 0;
2362 int consider_waiting;
2364 if (root == root->fs_info->extent_root)
2365 root = root->fs_info->tree_root;
2367 do_chunk_alloc(trans, root->fs_info->extent_root,
2368 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2369 CHUNK_ALLOC_NO_FORCE);
2371 delayed_refs = &trans->transaction->delayed_refs;
2372 INIT_LIST_HEAD(&cluster);
2374 consider_waiting = 0;
2375 spin_lock(&delayed_refs->lock);
2377 count = delayed_refs->num_entries * 2;
2381 if (!(run_all || run_most) &&
2382 delayed_refs->num_heads_ready < 64)
2386 * go find something we can process in the rbtree. We start at
2387 * the beginning of the tree, and then build a cluster
2388 * of refs to process starting at the first one we are able to
2391 delayed_start = delayed_refs->run_delayed_start;
2392 ret = btrfs_find_ref_cluster(trans, &cluster,
2393 delayed_refs->run_delayed_start);
2397 if (delayed_start >= delayed_refs->run_delayed_start) {
2398 if (consider_waiting == 0) {
2400 * btrfs_find_ref_cluster looped. let's do one
2401 * more cycle. if we don't run any delayed ref
2402 * during that cycle (because we can't because
2403 * all of them are blocked) and if the number of
2404 * refs doesn't change, we avoid busy waiting.
2406 consider_waiting = 1;
2407 num_refs = delayed_refs->num_entries;
2409 wait_for_more_refs(delayed_refs, num_refs);
2411 * after waiting, things have changed. we
2412 * dropped the lock and someone else might have
2413 * run some refs, built new clusters and so on.
2414 * therefore, we restart staleness detection.
2416 consider_waiting = 0;
2420 ret = run_clustered_refs(trans, root, &cluster);
2423 count -= min_t(unsigned long, ret, count);
2428 if (ret || delayed_refs->run_delayed_start == 0) {
2429 /* refs were run, let's reset staleness detection */
2430 consider_waiting = 0;
2435 node = rb_first(&delayed_refs->root);
2438 count = (unsigned long)-1;
2441 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2443 if (btrfs_delayed_ref_is_head(ref)) {
2444 struct btrfs_delayed_ref_head *head;
2446 head = btrfs_delayed_node_to_head(ref);
2447 atomic_inc(&ref->refs);
2449 spin_unlock(&delayed_refs->lock);
2451 * Mutex was contended, block until it's
2452 * released and try again
2454 mutex_lock(&head->mutex);
2455 mutex_unlock(&head->mutex);
2457 btrfs_put_delayed_ref(ref);
2461 node = rb_next(node);
2463 spin_unlock(&delayed_refs->lock);
2464 schedule_timeout(1);
2468 spin_unlock(&delayed_refs->lock);
2472 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2473 struct btrfs_root *root,
2474 u64 bytenr, u64 num_bytes, u64 flags,
2477 struct btrfs_delayed_extent_op *extent_op;
2480 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2484 extent_op->flags_to_set = flags;
2485 extent_op->update_flags = 1;
2486 extent_op->update_key = 0;
2487 extent_op->is_data = is_data ? 1 : 0;
2489 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2490 num_bytes, extent_op);
2496 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2497 struct btrfs_root *root,
2498 struct btrfs_path *path,
2499 u64 objectid, u64 offset, u64 bytenr)
2501 struct btrfs_delayed_ref_head *head;
2502 struct btrfs_delayed_ref_node *ref;
2503 struct btrfs_delayed_data_ref *data_ref;
2504 struct btrfs_delayed_ref_root *delayed_refs;
2505 struct rb_node *node;
2509 delayed_refs = &trans->transaction->delayed_refs;
2510 spin_lock(&delayed_refs->lock);
2511 head = btrfs_find_delayed_ref_head(trans, bytenr);
2515 if (!mutex_trylock(&head->mutex)) {
2516 atomic_inc(&head->node.refs);
2517 spin_unlock(&delayed_refs->lock);
2519 btrfs_release_path(path);
2522 * Mutex was contended, block until it's released and let
2525 mutex_lock(&head->mutex);
2526 mutex_unlock(&head->mutex);
2527 btrfs_put_delayed_ref(&head->node);
2531 node = rb_prev(&head->node.rb_node);
2535 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2537 if (ref->bytenr != bytenr)
2541 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2544 data_ref = btrfs_delayed_node_to_data_ref(ref);
2546 node = rb_prev(node);
2548 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2549 if (ref->bytenr == bytenr)
2553 if (data_ref->root != root->root_key.objectid ||
2554 data_ref->objectid != objectid || data_ref->offset != offset)
2559 mutex_unlock(&head->mutex);
2561 spin_unlock(&delayed_refs->lock);
2565 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2566 struct btrfs_root *root,
2567 struct btrfs_path *path,
2568 u64 objectid, u64 offset, u64 bytenr)
2570 struct btrfs_root *extent_root = root->fs_info->extent_root;
2571 struct extent_buffer *leaf;
2572 struct btrfs_extent_data_ref *ref;
2573 struct btrfs_extent_inline_ref *iref;
2574 struct btrfs_extent_item *ei;
2575 struct btrfs_key key;
2579 key.objectid = bytenr;
2580 key.offset = (u64)-1;
2581 key.type = BTRFS_EXTENT_ITEM_KEY;
2583 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2589 if (path->slots[0] == 0)
2593 leaf = path->nodes[0];
2594 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2596 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2600 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2601 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2602 if (item_size < sizeof(*ei)) {
2603 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2607 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2609 if (item_size != sizeof(*ei) +
2610 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2613 if (btrfs_extent_generation(leaf, ei) <=
2614 btrfs_root_last_snapshot(&root->root_item))
2617 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2618 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2619 BTRFS_EXTENT_DATA_REF_KEY)
2622 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2623 if (btrfs_extent_refs(leaf, ei) !=
2624 btrfs_extent_data_ref_count(leaf, ref) ||
2625 btrfs_extent_data_ref_root(leaf, ref) !=
2626 root->root_key.objectid ||
2627 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2628 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2636 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2637 struct btrfs_root *root,
2638 u64 objectid, u64 offset, u64 bytenr)
2640 struct btrfs_path *path;
2644 path = btrfs_alloc_path();
2649 ret = check_committed_ref(trans, root, path, objectid,
2651 if (ret && ret != -ENOENT)
2654 ret2 = check_delayed_ref(trans, root, path, objectid,
2656 } while (ret2 == -EAGAIN);
2658 if (ret2 && ret2 != -ENOENT) {
2663 if (ret != -ENOENT || ret2 != -ENOENT)
2666 btrfs_free_path(path);
2667 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2672 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2673 struct btrfs_root *root,
2674 struct extent_buffer *buf,
2675 int full_backref, int inc, int for_cow)
2682 struct btrfs_key key;
2683 struct btrfs_file_extent_item *fi;
2687 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2688 u64, u64, u64, u64, u64, u64, int);
2690 ref_root = btrfs_header_owner(buf);
2691 nritems = btrfs_header_nritems(buf);
2692 level = btrfs_header_level(buf);
2694 if (!root->ref_cows && level == 0)
2698 process_func = btrfs_inc_extent_ref;
2700 process_func = btrfs_free_extent;
2703 parent = buf->start;
2707 for (i = 0; i < nritems; i++) {
2709 btrfs_item_key_to_cpu(buf, &key, i);
2710 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2712 fi = btrfs_item_ptr(buf, i,
2713 struct btrfs_file_extent_item);
2714 if (btrfs_file_extent_type(buf, fi) ==
2715 BTRFS_FILE_EXTENT_INLINE)
2717 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2721 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2722 key.offset -= btrfs_file_extent_offset(buf, fi);
2723 ret = process_func(trans, root, bytenr, num_bytes,
2724 parent, ref_root, key.objectid,
2725 key.offset, for_cow);
2729 bytenr = btrfs_node_blockptr(buf, i);
2730 num_bytes = btrfs_level_size(root, level - 1);
2731 ret = process_func(trans, root, bytenr, num_bytes,
2732 parent, ref_root, level - 1, 0,
2744 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2745 struct extent_buffer *buf, int full_backref, int for_cow)
2747 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2750 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2751 struct extent_buffer *buf, int full_backref, int for_cow)
2753 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2756 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2757 struct btrfs_root *root,
2758 struct btrfs_path *path,
2759 struct btrfs_block_group_cache *cache)
2762 struct btrfs_root *extent_root = root->fs_info->extent_root;
2764 struct extent_buffer *leaf;
2766 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2771 leaf = path->nodes[0];
2772 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2773 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2774 btrfs_mark_buffer_dirty(leaf);
2775 btrfs_release_path(path);
2783 static struct btrfs_block_group_cache *
2784 next_block_group(struct btrfs_root *root,
2785 struct btrfs_block_group_cache *cache)
2787 struct rb_node *node;
2788 spin_lock(&root->fs_info->block_group_cache_lock);
2789 node = rb_next(&cache->cache_node);
2790 btrfs_put_block_group(cache);
2792 cache = rb_entry(node, struct btrfs_block_group_cache,
2794 btrfs_get_block_group(cache);
2797 spin_unlock(&root->fs_info->block_group_cache_lock);
2801 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2802 struct btrfs_trans_handle *trans,
2803 struct btrfs_path *path)
2805 struct btrfs_root *root = block_group->fs_info->tree_root;
2806 struct inode *inode = NULL;
2808 int dcs = BTRFS_DC_ERROR;
2814 * If this block group is smaller than 100 megs don't bother caching the
2817 if (block_group->key.offset < (100 * 1024 * 1024)) {
2818 spin_lock(&block_group->lock);
2819 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2820 spin_unlock(&block_group->lock);
2825 inode = lookup_free_space_inode(root, block_group, path);
2826 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2827 ret = PTR_ERR(inode);
2828 btrfs_release_path(path);
2832 if (IS_ERR(inode)) {
2836 if (block_group->ro)
2839 ret = create_free_space_inode(root, trans, block_group, path);
2845 /* We've already setup this transaction, go ahead and exit */
2846 if (block_group->cache_generation == trans->transid &&
2847 i_size_read(inode)) {
2848 dcs = BTRFS_DC_SETUP;
2853 * We want to set the generation to 0, that way if anything goes wrong
2854 * from here on out we know not to trust this cache when we load up next
2857 BTRFS_I(inode)->generation = 0;
2858 ret = btrfs_update_inode(trans, root, inode);
2861 if (i_size_read(inode) > 0) {
2862 ret = btrfs_truncate_free_space_cache(root, trans, path,
2868 spin_lock(&block_group->lock);
2869 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2870 /* We're not cached, don't bother trying to write stuff out */
2871 dcs = BTRFS_DC_WRITTEN;
2872 spin_unlock(&block_group->lock);
2875 spin_unlock(&block_group->lock);
2877 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2882 * Just to make absolutely sure we have enough space, we're going to
2883 * preallocate 12 pages worth of space for each block group. In
2884 * practice we ought to use at most 8, but we need extra space so we can
2885 * add our header and have a terminator between the extents and the
2889 num_pages *= PAGE_CACHE_SIZE;
2891 ret = btrfs_check_data_free_space(inode, num_pages);
2895 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2896 num_pages, num_pages,
2899 dcs = BTRFS_DC_SETUP;
2900 btrfs_free_reserved_data_space(inode, num_pages);
2905 btrfs_release_path(path);
2907 spin_lock(&block_group->lock);
2908 if (!ret && dcs == BTRFS_DC_SETUP)
2909 block_group->cache_generation = trans->transid;
2910 block_group->disk_cache_state = dcs;
2911 spin_unlock(&block_group->lock);
2916 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2917 struct btrfs_root *root)
2919 struct btrfs_block_group_cache *cache;
2921 struct btrfs_path *path;
2924 path = btrfs_alloc_path();
2930 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2932 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2934 cache = next_block_group(root, cache);
2942 err = cache_save_setup(cache, trans, path);
2943 last = cache->key.objectid + cache->key.offset;
2944 btrfs_put_block_group(cache);
2949 err = btrfs_run_delayed_refs(trans, root,
2954 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2956 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2957 btrfs_put_block_group(cache);
2963 cache = next_block_group(root, cache);
2972 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2973 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2975 last = cache->key.objectid + cache->key.offset;
2977 err = write_one_cache_group(trans, root, path, cache);
2979 btrfs_put_block_group(cache);
2984 * I don't think this is needed since we're just marking our
2985 * preallocated extent as written, but just in case it can't
2989 err = btrfs_run_delayed_refs(trans, root,
2994 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2997 * Really this shouldn't happen, but it could if we
2998 * couldn't write the entire preallocated extent and
2999 * splitting the extent resulted in a new block.
3002 btrfs_put_block_group(cache);
3005 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3007 cache = next_block_group(root, cache);
3016 btrfs_write_out_cache(root, trans, cache, path);
3019 * If we didn't have an error then the cache state is still
3020 * NEED_WRITE, so we can set it to WRITTEN.
3022 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3023 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3024 last = cache->key.objectid + cache->key.offset;
3025 btrfs_put_block_group(cache);
3028 btrfs_free_path(path);
3032 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3034 struct btrfs_block_group_cache *block_group;
3037 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3038 if (!block_group || block_group->ro)
3041 btrfs_put_block_group(block_group);
3045 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3046 u64 total_bytes, u64 bytes_used,
3047 struct btrfs_space_info **space_info)
3049 struct btrfs_space_info *found;
3053 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3054 BTRFS_BLOCK_GROUP_RAID10))
3059 found = __find_space_info(info, flags);
3061 spin_lock(&found->lock);
3062 found->total_bytes += total_bytes;
3063 found->disk_total += total_bytes * factor;
3064 found->bytes_used += bytes_used;
3065 found->disk_used += bytes_used * factor;
3067 spin_unlock(&found->lock);
3068 *space_info = found;
3071 found = kzalloc(sizeof(*found), GFP_NOFS);
3075 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3076 INIT_LIST_HEAD(&found->block_groups[i]);
3077 init_rwsem(&found->groups_sem);
3078 spin_lock_init(&found->lock);
3079 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3080 found->total_bytes = total_bytes;
3081 found->disk_total = total_bytes * factor;
3082 found->bytes_used = bytes_used;
3083 found->disk_used = bytes_used * factor;
3084 found->bytes_pinned = 0;
3085 found->bytes_reserved = 0;
3086 found->bytes_readonly = 0;
3087 found->bytes_may_use = 0;
3089 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3090 found->chunk_alloc = 0;
3092 init_waitqueue_head(&found->wait);
3093 *space_info = found;
3094 list_add_rcu(&found->list, &info->space_info);
3098 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3100 u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
3102 /* chunk -> extended profile */
3103 if (extra_flags == 0)
3104 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3106 if (flags & BTRFS_BLOCK_GROUP_DATA)
3107 fs_info->avail_data_alloc_bits |= extra_flags;
3108 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3109 fs_info->avail_metadata_alloc_bits |= extra_flags;
3110 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3111 fs_info->avail_system_alloc_bits |= extra_flags;
3115 * @flags: available profiles in extended format (see ctree.h)
3117 * Returns reduced profile in chunk format. If profile changing is in
3118 * progress (either running or paused) picks the target profile (if it's
3119 * already available), otherwise falls back to plain reducing.
3121 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3124 * we add in the count of missing devices because we want
3125 * to make sure that any RAID levels on a degraded FS
3126 * continue to be honored.
3128 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3129 root->fs_info->fs_devices->missing_devices;
3131 /* pick restriper's target profile if it's available */
3132 spin_lock(&root->fs_info->balance_lock);
3133 if (root->fs_info->balance_ctl) {
3134 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3137 if ((flags & BTRFS_BLOCK_GROUP_DATA) &&
3138 (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3139 (flags & bctl->data.target)) {
3140 tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3141 } else if ((flags & BTRFS_BLOCK_GROUP_SYSTEM) &&
3142 (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3143 (flags & bctl->sys.target)) {
3144 tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3145 } else if ((flags & BTRFS_BLOCK_GROUP_METADATA) &&
3146 (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3147 (flags & bctl->meta.target)) {
3148 tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3152 spin_unlock(&root->fs_info->balance_lock);
3157 spin_unlock(&root->fs_info->balance_lock);
3159 if (num_devices == 1)
3160 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3161 if (num_devices < 4)
3162 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3164 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3165 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3166 BTRFS_BLOCK_GROUP_RAID10))) {
3167 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3170 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3171 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3172 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3175 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3176 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3177 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3178 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3179 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3183 /* extended -> chunk profile */
3184 flags &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3188 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3190 if (flags & BTRFS_BLOCK_GROUP_DATA)
3191 flags |= root->fs_info->avail_data_alloc_bits;
3192 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3193 flags |= root->fs_info->avail_system_alloc_bits;
3194 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3195 flags |= root->fs_info->avail_metadata_alloc_bits;
3197 return btrfs_reduce_alloc_profile(root, flags);
3200 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3205 flags = BTRFS_BLOCK_GROUP_DATA;
3206 else if (root == root->fs_info->chunk_root)
3207 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3209 flags = BTRFS_BLOCK_GROUP_METADATA;
3211 return get_alloc_profile(root, flags);
3214 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3216 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3217 BTRFS_BLOCK_GROUP_DATA);
3221 * This will check the space that the inode allocates from to make sure we have
3222 * enough space for bytes.
3224 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3226 struct btrfs_space_info *data_sinfo;
3227 struct btrfs_root *root = BTRFS_I(inode)->root;
3229 int ret = 0, committed = 0, alloc_chunk = 1;
3231 /* make sure bytes are sectorsize aligned */
3232 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3234 if (root == root->fs_info->tree_root ||
3235 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3240 data_sinfo = BTRFS_I(inode)->space_info;
3245 /* make sure we have enough space to handle the data first */
3246 spin_lock(&data_sinfo->lock);
3247 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3248 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3249 data_sinfo->bytes_may_use;
3251 if (used + bytes > data_sinfo->total_bytes) {
3252 struct btrfs_trans_handle *trans;
3255 * if we don't have enough free bytes in this space then we need
3256 * to alloc a new chunk.
3258 if (!data_sinfo->full && alloc_chunk) {
3261 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3262 spin_unlock(&data_sinfo->lock);
3264 alloc_target = btrfs_get_alloc_profile(root, 1);
3265 trans = btrfs_join_transaction(root);
3267 return PTR_ERR(trans);
3269 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3270 bytes + 2 * 1024 * 1024,
3272 CHUNK_ALLOC_NO_FORCE);
3273 btrfs_end_transaction(trans, root);
3282 btrfs_set_inode_space_info(root, inode);
3283 data_sinfo = BTRFS_I(inode)->space_info;
3289 * If we have less pinned bytes than we want to allocate then
3290 * don't bother committing the transaction, it won't help us.
3292 if (data_sinfo->bytes_pinned < bytes)
3294 spin_unlock(&data_sinfo->lock);
3296 /* commit the current transaction and try again */
3299 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3301 trans = btrfs_join_transaction(root);
3303 return PTR_ERR(trans);
3304 ret = btrfs_commit_transaction(trans, root);
3312 data_sinfo->bytes_may_use += bytes;
3313 spin_unlock(&data_sinfo->lock);
3319 * Called if we need to clear a data reservation for this inode.
3321 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3323 struct btrfs_root *root = BTRFS_I(inode)->root;
3324 struct btrfs_space_info *data_sinfo;
3326 /* make sure bytes are sectorsize aligned */
3327 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3329 data_sinfo = BTRFS_I(inode)->space_info;
3330 spin_lock(&data_sinfo->lock);
3331 data_sinfo->bytes_may_use -= bytes;
3332 spin_unlock(&data_sinfo->lock);
3335 static void force_metadata_allocation(struct btrfs_fs_info *info)
3337 struct list_head *head = &info->space_info;
3338 struct btrfs_space_info *found;
3341 list_for_each_entry_rcu(found, head, list) {
3342 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3343 found->force_alloc = CHUNK_ALLOC_FORCE;
3348 static int should_alloc_chunk(struct btrfs_root *root,
3349 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3352 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3353 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3354 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3357 if (force == CHUNK_ALLOC_FORCE)
3361 * We need to take into account the global rsv because for all intents
3362 * and purposes it's used space. Don't worry about locking the
3363 * global_rsv, it doesn't change except when the transaction commits.
3365 num_allocated += global_rsv->size;
3368 * in limited mode, we want to have some free space up to
3369 * about 1% of the FS size.
3371 if (force == CHUNK_ALLOC_LIMITED) {
3372 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3373 thresh = max_t(u64, 64 * 1024 * 1024,
3374 div_factor_fine(thresh, 1));
3376 if (num_bytes - num_allocated < thresh)
3379 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3381 /* 256MB or 2% of the FS */
3382 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3384 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3389 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3390 struct btrfs_root *extent_root, u64 alloc_bytes,
3391 u64 flags, int force)
3393 struct btrfs_space_info *space_info;
3394 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3395 int wait_for_alloc = 0;
3398 BUG_ON(!profile_is_valid(flags, 0));
3400 space_info = __find_space_info(extent_root->fs_info, flags);
3402 ret = update_space_info(extent_root->fs_info, flags,
3406 BUG_ON(!space_info);
3409 spin_lock(&space_info->lock);
3410 if (space_info->force_alloc)
3411 force = space_info->force_alloc;
3412 if (space_info->full) {
3413 spin_unlock(&space_info->lock);
3417 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3418 spin_unlock(&space_info->lock);
3420 } else if (space_info->chunk_alloc) {
3423 space_info->chunk_alloc = 1;
3426 spin_unlock(&space_info->lock);
3428 mutex_lock(&fs_info->chunk_mutex);
3431 * The chunk_mutex is held throughout the entirety of a chunk
3432 * allocation, so once we've acquired the chunk_mutex we know that the
3433 * other guy is done and we need to recheck and see if we should
3436 if (wait_for_alloc) {
3437 mutex_unlock(&fs_info->chunk_mutex);
3443 * If we have mixed data/metadata chunks we want to make sure we keep
3444 * allocating mixed chunks instead of individual chunks.
3446 if (btrfs_mixed_space_info(space_info))
3447 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3450 * if we're doing a data chunk, go ahead and make sure that
3451 * we keep a reasonable number of metadata chunks allocated in the
3454 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3455 fs_info->data_chunk_allocations++;
3456 if (!(fs_info->data_chunk_allocations %
3457 fs_info->metadata_ratio))
3458 force_metadata_allocation(fs_info);
3461 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3462 if (ret < 0 && ret != -ENOSPC)
3465 spin_lock(&space_info->lock);
3467 space_info->full = 1;
3471 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3472 space_info->chunk_alloc = 0;
3473 spin_unlock(&space_info->lock);
3475 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3480 * shrink metadata reservation for delalloc
3482 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3485 struct btrfs_block_rsv *block_rsv;
3486 struct btrfs_space_info *space_info;
3487 struct btrfs_trans_handle *trans;
3492 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3494 unsigned long progress;
3496 trans = (struct btrfs_trans_handle *)current->journal_info;
3497 block_rsv = &root->fs_info->delalloc_block_rsv;
3498 space_info = block_rsv->space_info;
3501 reserved = space_info->bytes_may_use;
3502 progress = space_info->reservation_progress;
3508 if (root->fs_info->delalloc_bytes == 0) {
3511 btrfs_wait_ordered_extents(root, 0, 0);
3515 max_reclaim = min(reserved, to_reclaim);
3516 nr_pages = max_t(unsigned long, nr_pages,
3517 max_reclaim >> PAGE_CACHE_SHIFT);
3518 while (loops < 1024) {
3519 /* have the flusher threads jump in and do some IO */
3521 nr_pages = min_t(unsigned long, nr_pages,
3522 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3523 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3524 WB_REASON_FS_FREE_SPACE);
3526 spin_lock(&space_info->lock);
3527 if (reserved > space_info->bytes_may_use)
3528 reclaimed += reserved - space_info->bytes_may_use;
3529 reserved = space_info->bytes_may_use;
3530 spin_unlock(&space_info->lock);
3534 if (reserved == 0 || reclaimed >= max_reclaim)
3537 if (trans && trans->transaction->blocked)
3540 if (wait_ordered && !trans) {
3541 btrfs_wait_ordered_extents(root, 0, 0);
3543 time_left = schedule_timeout_interruptible(1);
3545 /* We were interrupted, exit */
3550 /* we've kicked the IO a few times, if anything has been freed,
3551 * exit. There is no sense in looping here for a long time
3552 * when we really need to commit the transaction, or there are
3553 * just too many writers without enough free space
3558 if (progress != space_info->reservation_progress)
3564 return reclaimed >= to_reclaim;
3568 * maybe_commit_transaction - possibly commit the transaction if its ok to
3569 * @root - the root we're allocating for
3570 * @bytes - the number of bytes we want to reserve
3571 * @force - force the commit
3573 * This will check to make sure that committing the transaction will actually
3574 * get us somewhere and then commit the transaction if it does. Otherwise it
3575 * will return -ENOSPC.
3577 static int may_commit_transaction(struct btrfs_root *root,
3578 struct btrfs_space_info *space_info,
3579 u64 bytes, int force)
3581 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3582 struct btrfs_trans_handle *trans;
3584 trans = (struct btrfs_trans_handle *)current->journal_info;
3591 /* See if there is enough pinned space to make this reservation */
3592 spin_lock(&space_info->lock);
3593 if (space_info->bytes_pinned >= bytes) {
3594 spin_unlock(&space_info->lock);
3597 spin_unlock(&space_info->lock);
3600 * See if there is some space in the delayed insertion reservation for
3603 if (space_info != delayed_rsv->space_info)
3606 spin_lock(&delayed_rsv->lock);
3607 if (delayed_rsv->size < bytes) {
3608 spin_unlock(&delayed_rsv->lock);
3611 spin_unlock(&delayed_rsv->lock);
3614 trans = btrfs_join_transaction(root);
3618 return btrfs_commit_transaction(trans, root);
3622 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3623 * @root - the root we're allocating for
3624 * @block_rsv - the block_rsv we're allocating for
3625 * @orig_bytes - the number of bytes we want
3626 * @flush - wether or not we can flush to make our reservation
3628 * This will reserve orgi_bytes number of bytes from the space info associated
3629 * with the block_rsv. If there is not enough space it will make an attempt to
3630 * flush out space to make room. It will do this by flushing delalloc if
3631 * possible or committing the transaction. If flush is 0 then no attempts to
3632 * regain reservations will be made and this will fail if there is not enough
3635 static int reserve_metadata_bytes(struct btrfs_root *root,
3636 struct btrfs_block_rsv *block_rsv,
3637 u64 orig_bytes, int flush)
3639 struct btrfs_space_info *space_info = block_rsv->space_info;
3641 u64 num_bytes = orig_bytes;
3644 bool committed = false;
3645 bool flushing = false;
3646 bool wait_ordered = false;
3650 spin_lock(&space_info->lock);
3652 * We only want to wait if somebody other than us is flushing and we are
3653 * actually alloed to flush.
3655 while (flush && !flushing && space_info->flush) {
3656 spin_unlock(&space_info->lock);
3658 * If we have a trans handle we can't wait because the flusher
3659 * may have to commit the transaction, which would mean we would
3660 * deadlock since we are waiting for the flusher to finish, but
3661 * hold the current transaction open.
3663 if (current->journal_info)
3665 ret = wait_event_interruptible(space_info->wait,
3666 !space_info->flush);
3667 /* Must have been interrupted, return */
3671 spin_lock(&space_info->lock);
3675 used = space_info->bytes_used + space_info->bytes_reserved +
3676 space_info->bytes_pinned + space_info->bytes_readonly +
3677 space_info->bytes_may_use;
3680 * The idea here is that we've not already over-reserved the block group
3681 * then we can go ahead and save our reservation first and then start
3682 * flushing if we need to. Otherwise if we've already overcommitted
3683 * lets start flushing stuff first and then come back and try to make
3686 if (used <= space_info->total_bytes) {
3687 if (used + orig_bytes <= space_info->total_bytes) {
3688 space_info->bytes_may_use += orig_bytes;
3692 * Ok set num_bytes to orig_bytes since we aren't
3693 * overocmmitted, this way we only try and reclaim what
3696 num_bytes = orig_bytes;
3700 * Ok we're over committed, set num_bytes to the overcommitted
3701 * amount plus the amount of bytes that we need for this
3704 wait_ordered = true;
3705 num_bytes = used - space_info->total_bytes +
3706 (orig_bytes * (retries + 1));
3710 u64 profile = btrfs_get_alloc_profile(root, 0);
3714 * If we have a lot of space that's pinned, don't bother doing
3715 * the overcommit dance yet and just commit the transaction.
3717 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3719 if (space_info->bytes_pinned >= avail && flush && !committed) {
3720 space_info->flush = 1;
3722 spin_unlock(&space_info->lock);
3723 ret = may_commit_transaction(root, space_info,
3731 spin_lock(&root->fs_info->free_chunk_lock);
3732 avail = root->fs_info->free_chunk_space;
3735 * If we have dup, raid1 or raid10 then only half of the free
3736 * space is actually useable.
3738 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3739 BTRFS_BLOCK_GROUP_RAID1 |
3740 BTRFS_BLOCK_GROUP_RAID10))
3744 * If we aren't flushing don't let us overcommit too much, say
3745 * 1/8th of the space. If we can flush, let it overcommit up to
3752 spin_unlock(&root->fs_info->free_chunk_lock);
3754 if (used + num_bytes < space_info->total_bytes + avail) {
3755 space_info->bytes_may_use += orig_bytes;
3758 wait_ordered = true;
3763 * Couldn't make our reservation, save our place so while we're trying
3764 * to reclaim space we can actually use it instead of somebody else
3765 * stealing it from us.
3769 space_info->flush = 1;
3772 spin_unlock(&space_info->lock);
3778 * We do synchronous shrinking since we don't actually unreserve
3779 * metadata until after the IO is completed.
3781 ret = shrink_delalloc(root, num_bytes, wait_ordered);
3788 * So if we were overcommitted it's possible that somebody else flushed
3789 * out enough space and we simply didn't have enough space to reclaim,
3790 * so go back around and try again.
3793 wait_ordered = true;
3802 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3810 spin_lock(&space_info->lock);
3811 space_info->flush = 0;
3812 wake_up_all(&space_info->wait);
3813 spin_unlock(&space_info->lock);
3818 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3819 struct btrfs_root *root)
3821 struct btrfs_block_rsv *block_rsv = NULL;
3823 if (root->ref_cows || root == root->fs_info->csum_root)
3824 block_rsv = trans->block_rsv;
3827 block_rsv = root->block_rsv;
3830 block_rsv = &root->fs_info->empty_block_rsv;
3835 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3839 spin_lock(&block_rsv->lock);
3840 if (block_rsv->reserved >= num_bytes) {
3841 block_rsv->reserved -= num_bytes;
3842 if (block_rsv->reserved < block_rsv->size)
3843 block_rsv->full = 0;
3846 spin_unlock(&block_rsv->lock);
3850 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3851 u64 num_bytes, int update_size)
3853 spin_lock(&block_rsv->lock);
3854 block_rsv->reserved += num_bytes;
3856 block_rsv->size += num_bytes;
3857 else if (block_rsv->reserved >= block_rsv->size)
3858 block_rsv->full = 1;
3859 spin_unlock(&block_rsv->lock);
3862 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3863 struct btrfs_block_rsv *dest, u64 num_bytes)
3865 struct btrfs_space_info *space_info = block_rsv->space_info;
3867 spin_lock(&block_rsv->lock);
3868 if (num_bytes == (u64)-1)
3869 num_bytes = block_rsv->size;
3870 block_rsv->size -= num_bytes;
3871 if (block_rsv->reserved >= block_rsv->size) {
3872 num_bytes = block_rsv->reserved - block_rsv->size;
3873 block_rsv->reserved = block_rsv->size;
3874 block_rsv->full = 1;
3878 spin_unlock(&block_rsv->lock);
3880 if (num_bytes > 0) {
3882 spin_lock(&dest->lock);
3886 bytes_to_add = dest->size - dest->reserved;
3887 bytes_to_add = min(num_bytes, bytes_to_add);
3888 dest->reserved += bytes_to_add;
3889 if (dest->reserved >= dest->size)
3891 num_bytes -= bytes_to_add;
3893 spin_unlock(&dest->lock);
3896 spin_lock(&space_info->lock);
3897 space_info->bytes_may_use -= num_bytes;
3898 space_info->reservation_progress++;
3899 spin_unlock(&space_info->lock);
3904 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3905 struct btrfs_block_rsv *dst, u64 num_bytes)
3909 ret = block_rsv_use_bytes(src, num_bytes);
3913 block_rsv_add_bytes(dst, num_bytes, 1);
3917 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3919 memset(rsv, 0, sizeof(*rsv));
3920 spin_lock_init(&rsv->lock);
3923 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3925 struct btrfs_block_rsv *block_rsv;
3926 struct btrfs_fs_info *fs_info = root->fs_info;
3928 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3932 btrfs_init_block_rsv(block_rsv);
3933 block_rsv->space_info = __find_space_info(fs_info,
3934 BTRFS_BLOCK_GROUP_METADATA);
3938 void btrfs_free_block_rsv(struct btrfs_root *root,
3939 struct btrfs_block_rsv *rsv)
3941 btrfs_block_rsv_release(root, rsv, (u64)-1);
3945 static inline int __block_rsv_add(struct btrfs_root *root,
3946 struct btrfs_block_rsv *block_rsv,
3947 u64 num_bytes, int flush)
3954 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3956 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3963 int btrfs_block_rsv_add(struct btrfs_root *root,
3964 struct btrfs_block_rsv *block_rsv,
3967 return __block_rsv_add(root, block_rsv, num_bytes, 1);
3970 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3971 struct btrfs_block_rsv *block_rsv,
3974 return __block_rsv_add(root, block_rsv, num_bytes, 0);
3977 int btrfs_block_rsv_check(struct btrfs_root *root,
3978 struct btrfs_block_rsv *block_rsv, int min_factor)
3986 spin_lock(&block_rsv->lock);
3987 num_bytes = div_factor(block_rsv->size, min_factor);
3988 if (block_rsv->reserved >= num_bytes)
3990 spin_unlock(&block_rsv->lock);
3995 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
3996 struct btrfs_block_rsv *block_rsv,
3997 u64 min_reserved, int flush)
4005 spin_lock(&block_rsv->lock);
4006 num_bytes = min_reserved;
4007 if (block_rsv->reserved >= num_bytes)
4010 num_bytes -= block_rsv->reserved;
4011 spin_unlock(&block_rsv->lock);
4016 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4018 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4025 int btrfs_block_rsv_refill(struct btrfs_root *root,
4026 struct btrfs_block_rsv *block_rsv,
4029 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4032 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4033 struct btrfs_block_rsv *block_rsv,
4036 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4039 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4040 struct btrfs_block_rsv *dst_rsv,
4043 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4046 void btrfs_block_rsv_release(struct btrfs_root *root,
4047 struct btrfs_block_rsv *block_rsv,
4050 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4051 if (global_rsv->full || global_rsv == block_rsv ||
4052 block_rsv->space_info != global_rsv->space_info)
4054 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
4058 * helper to calculate size of global block reservation.
4059 * the desired value is sum of space used by extent tree,
4060 * checksum tree and root tree
4062 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4064 struct btrfs_space_info *sinfo;
4068 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4070 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4071 spin_lock(&sinfo->lock);
4072 data_used = sinfo->bytes_used;
4073 spin_unlock(&sinfo->lock);
4075 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4076 spin_lock(&sinfo->lock);
4077 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4079 meta_used = sinfo->bytes_used;
4080 spin_unlock(&sinfo->lock);
4082 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4084 num_bytes += div64_u64(data_used + meta_used, 50);
4086 if (num_bytes * 3 > meta_used)
4087 num_bytes = div64_u64(meta_used, 3);
4089 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4092 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4094 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4095 struct btrfs_space_info *sinfo = block_rsv->space_info;
4098 num_bytes = calc_global_metadata_size(fs_info);
4100 spin_lock(&block_rsv->lock);
4101 spin_lock(&sinfo->lock);
4103 block_rsv->size = num_bytes;
4105 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4106 sinfo->bytes_reserved + sinfo->bytes_readonly +
4107 sinfo->bytes_may_use;
4109 if (sinfo->total_bytes > num_bytes) {
4110 num_bytes = sinfo->total_bytes - num_bytes;
4111 block_rsv->reserved += num_bytes;
4112 sinfo->bytes_may_use += num_bytes;
4115 if (block_rsv->reserved >= block_rsv->size) {
4116 num_bytes = block_rsv->reserved - block_rsv->size;
4117 sinfo->bytes_may_use -= num_bytes;
4118 sinfo->reservation_progress++;
4119 block_rsv->reserved = block_rsv->size;
4120 block_rsv->full = 1;
4123 spin_unlock(&sinfo->lock);
4124 spin_unlock(&block_rsv->lock);
4127 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4129 struct btrfs_space_info *space_info;
4131 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4132 fs_info->chunk_block_rsv.space_info = space_info;
4134 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4135 fs_info->global_block_rsv.space_info = space_info;
4136 fs_info->delalloc_block_rsv.space_info = space_info;
4137 fs_info->trans_block_rsv.space_info = space_info;
4138 fs_info->empty_block_rsv.space_info = space_info;
4139 fs_info->delayed_block_rsv.space_info = space_info;
4141 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4142 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4143 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4144 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4145 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4147 update_global_block_rsv(fs_info);
4150 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4152 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
4153 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4154 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4155 WARN_ON(fs_info->trans_block_rsv.size > 0);
4156 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4157 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4158 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4159 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4160 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4163 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4164 struct btrfs_root *root)
4166 if (!trans->bytes_reserved)
4169 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4170 trans->bytes_reserved = 0;
4173 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4174 struct inode *inode)
4176 struct btrfs_root *root = BTRFS_I(inode)->root;
4177 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4178 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4181 * We need to hold space in order to delete our orphan item once we've
4182 * added it, so this takes the reservation so we can release it later
4183 * when we are truly done with the orphan item.
4185 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4186 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4189 void btrfs_orphan_release_metadata(struct inode *inode)
4191 struct btrfs_root *root = BTRFS_I(inode)->root;
4192 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4193 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4196 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4197 struct btrfs_pending_snapshot *pending)
4199 struct btrfs_root *root = pending->root;
4200 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4201 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4203 * two for root back/forward refs, two for directory entries
4204 * and one for root of the snapshot.
4206 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4207 dst_rsv->space_info = src_rsv->space_info;
4208 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4212 * drop_outstanding_extent - drop an outstanding extent
4213 * @inode: the inode we're dropping the extent for
4215 * This is called when we are freeing up an outstanding extent, either called
4216 * after an error or after an extent is written. This will return the number of
4217 * reserved extents that need to be freed. This must be called with
4218 * BTRFS_I(inode)->lock held.
4220 static unsigned drop_outstanding_extent(struct inode *inode)
4222 unsigned drop_inode_space = 0;
4223 unsigned dropped_extents = 0;
4225 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4226 BTRFS_I(inode)->outstanding_extents--;
4228 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4229 BTRFS_I(inode)->delalloc_meta_reserved) {
4230 drop_inode_space = 1;
4231 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4235 * If we have more or the same amount of outsanding extents than we have
4236 * reserved then we need to leave the reserved extents count alone.
4238 if (BTRFS_I(inode)->outstanding_extents >=
4239 BTRFS_I(inode)->reserved_extents)
4240 return drop_inode_space;
4242 dropped_extents = BTRFS_I(inode)->reserved_extents -
4243 BTRFS_I(inode)->outstanding_extents;
4244 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4245 return dropped_extents + drop_inode_space;
4249 * calc_csum_metadata_size - return the amount of metada space that must be
4250 * reserved/free'd for the given bytes.
4251 * @inode: the inode we're manipulating
4252 * @num_bytes: the number of bytes in question
4253 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4255 * This adjusts the number of csum_bytes in the inode and then returns the
4256 * correct amount of metadata that must either be reserved or freed. We
4257 * calculate how many checksums we can fit into one leaf and then divide the
4258 * number of bytes that will need to be checksumed by this value to figure out
4259 * how many checksums will be required. If we are adding bytes then the number
4260 * may go up and we will return the number of additional bytes that must be
4261 * reserved. If it is going down we will return the number of bytes that must
4264 * This must be called with BTRFS_I(inode)->lock held.
4266 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4269 struct btrfs_root *root = BTRFS_I(inode)->root;
4271 int num_csums_per_leaf;
4275 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4276 BTRFS_I(inode)->csum_bytes == 0)
4279 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4281 BTRFS_I(inode)->csum_bytes += num_bytes;
4283 BTRFS_I(inode)->csum_bytes -= num_bytes;
4284 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4285 num_csums_per_leaf = (int)div64_u64(csum_size,
4286 sizeof(struct btrfs_csum_item) +
4287 sizeof(struct btrfs_disk_key));
4288 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4289 num_csums = num_csums + num_csums_per_leaf - 1;
4290 num_csums = num_csums / num_csums_per_leaf;
4292 old_csums = old_csums + num_csums_per_leaf - 1;
4293 old_csums = old_csums / num_csums_per_leaf;
4295 /* No change, no need to reserve more */
4296 if (old_csums == num_csums)
4300 return btrfs_calc_trans_metadata_size(root,
4301 num_csums - old_csums);
4303 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4306 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4308 struct btrfs_root *root = BTRFS_I(inode)->root;
4309 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4312 unsigned nr_extents = 0;
4313 int extra_reserve = 0;
4317 /* Need to be holding the i_mutex here if we aren't free space cache */
4318 if (btrfs_is_free_space_inode(root, inode))
4321 WARN_ON(!mutex_is_locked(&inode->i_mutex));
4323 if (flush && btrfs_transaction_in_commit(root->fs_info))
4324 schedule_timeout(1);
4326 num_bytes = ALIGN(num_bytes, root->sectorsize);
4328 spin_lock(&BTRFS_I(inode)->lock);
4329 BTRFS_I(inode)->outstanding_extents++;
4331 if (BTRFS_I(inode)->outstanding_extents >
4332 BTRFS_I(inode)->reserved_extents)
4333 nr_extents = BTRFS_I(inode)->outstanding_extents -
4334 BTRFS_I(inode)->reserved_extents;
4337 * Add an item to reserve for updating the inode when we complete the
4340 if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4345 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4346 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4347 csum_bytes = BTRFS_I(inode)->csum_bytes;
4348 spin_unlock(&BTRFS_I(inode)->lock);
4350 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4355 spin_lock(&BTRFS_I(inode)->lock);
4356 dropped = drop_outstanding_extent(inode);
4358 * If the inodes csum_bytes is the same as the original
4359 * csum_bytes then we know we haven't raced with any free()ers
4360 * so we can just reduce our inodes csum bytes and carry on.
4361 * Otherwise we have to do the normal free thing to account for
4362 * the case that the free side didn't free up its reserve
4363 * because of this outstanding reservation.
4365 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4366 calc_csum_metadata_size(inode, num_bytes, 0);
4368 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4369 spin_unlock(&BTRFS_I(inode)->lock);
4371 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4374 btrfs_block_rsv_release(root, block_rsv, to_free);
4378 spin_lock(&BTRFS_I(inode)->lock);
4379 if (extra_reserve) {
4380 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4383 BTRFS_I(inode)->reserved_extents += nr_extents;
4384 spin_unlock(&BTRFS_I(inode)->lock);
4386 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4392 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4393 * @inode: the inode to release the reservation for
4394 * @num_bytes: the number of bytes we're releasing
4396 * This will release the metadata reservation for an inode. This can be called
4397 * once we complete IO for a given set of bytes to release their metadata
4400 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4402 struct btrfs_root *root = BTRFS_I(inode)->root;
4406 num_bytes = ALIGN(num_bytes, root->sectorsize);
4407 spin_lock(&BTRFS_I(inode)->lock);
4408 dropped = drop_outstanding_extent(inode);
4410 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4411 spin_unlock(&BTRFS_I(inode)->lock);
4413 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4415 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4420 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4421 * @inode: inode we're writing to
4422 * @num_bytes: the number of bytes we want to allocate
4424 * This will do the following things
4426 * o reserve space in the data space info for num_bytes
4427 * o reserve space in the metadata space info based on number of outstanding
4428 * extents and how much csums will be needed
4429 * o add to the inodes ->delalloc_bytes
4430 * o add it to the fs_info's delalloc inodes list.
4432 * This will return 0 for success and -ENOSPC if there is no space left.
4434 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4438 ret = btrfs_check_data_free_space(inode, num_bytes);
4442 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4444 btrfs_free_reserved_data_space(inode, num_bytes);
4452 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4453 * @inode: inode we're releasing space for
4454 * @num_bytes: the number of bytes we want to free up
4456 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4457 * called in the case that we don't need the metadata AND data reservations
4458 * anymore. So if there is an error or we insert an inline extent.
4460 * This function will release the metadata space that was not used and will
4461 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4462 * list if there are no delalloc bytes left.
4464 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4466 btrfs_delalloc_release_metadata(inode, num_bytes);
4467 btrfs_free_reserved_data_space(inode, num_bytes);
4470 static int update_block_group(struct btrfs_trans_handle *trans,
4471 struct btrfs_root *root,
4472 u64 bytenr, u64 num_bytes, int alloc)
4474 struct btrfs_block_group_cache *cache = NULL;
4475 struct btrfs_fs_info *info = root->fs_info;
4476 u64 total = num_bytes;
4481 /* block accounting for super block */
4482 spin_lock(&info->delalloc_lock);
4483 old_val = btrfs_super_bytes_used(info->super_copy);
4485 old_val += num_bytes;
4487 old_val -= num_bytes;
4488 btrfs_set_super_bytes_used(info->super_copy, old_val);
4489 spin_unlock(&info->delalloc_lock);
4492 cache = btrfs_lookup_block_group(info, bytenr);
4495 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4496 BTRFS_BLOCK_GROUP_RAID1 |
4497 BTRFS_BLOCK_GROUP_RAID10))
4502 * If this block group has free space cache written out, we
4503 * need to make sure to load it if we are removing space. This
4504 * is because we need the unpinning stage to actually add the
4505 * space back to the block group, otherwise we will leak space.
4507 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4508 cache_block_group(cache, trans, NULL, 1);
4510 byte_in_group = bytenr - cache->key.objectid;
4511 WARN_ON(byte_in_group > cache->key.offset);
4513 spin_lock(&cache->space_info->lock);
4514 spin_lock(&cache->lock);
4516 if (btrfs_test_opt(root, SPACE_CACHE) &&
4517 cache->disk_cache_state < BTRFS_DC_CLEAR)
4518 cache->disk_cache_state = BTRFS_DC_CLEAR;
4521 old_val = btrfs_block_group_used(&cache->item);
4522 num_bytes = min(total, cache->key.offset - byte_in_group);
4524 old_val += num_bytes;
4525 btrfs_set_block_group_used(&cache->item, old_val);
4526 cache->reserved -= num_bytes;
4527 cache->space_info->bytes_reserved -= num_bytes;
4528 cache->space_info->bytes_used += num_bytes;
4529 cache->space_info->disk_used += num_bytes * factor;
4530 spin_unlock(&cache->lock);
4531 spin_unlock(&cache->space_info->lock);
4533 old_val -= num_bytes;
4534 btrfs_set_block_group_used(&cache->item, old_val);
4535 cache->pinned += num_bytes;
4536 cache->space_info->bytes_pinned += num_bytes;
4537 cache->space_info->bytes_used -= num_bytes;
4538 cache->space_info->disk_used -= num_bytes * factor;
4539 spin_unlock(&cache->lock);
4540 spin_unlock(&cache->space_info->lock);
4542 set_extent_dirty(info->pinned_extents,
4543 bytenr, bytenr + num_bytes - 1,
4544 GFP_NOFS | __GFP_NOFAIL);
4546 btrfs_put_block_group(cache);
4548 bytenr += num_bytes;
4553 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4555 struct btrfs_block_group_cache *cache;
4558 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4562 bytenr = cache->key.objectid;
4563 btrfs_put_block_group(cache);
4568 static int pin_down_extent(struct btrfs_root *root,
4569 struct btrfs_block_group_cache *cache,
4570 u64 bytenr, u64 num_bytes, int reserved)
4572 spin_lock(&cache->space_info->lock);
4573 spin_lock(&cache->lock);
4574 cache->pinned += num_bytes;
4575 cache->space_info->bytes_pinned += num_bytes;
4577 cache->reserved -= num_bytes;
4578 cache->space_info->bytes_reserved -= num_bytes;
4580 spin_unlock(&cache->lock);
4581 spin_unlock(&cache->space_info->lock);
4583 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4584 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4589 * this function must be called within transaction
4591 int btrfs_pin_extent(struct btrfs_root *root,
4592 u64 bytenr, u64 num_bytes, int reserved)
4594 struct btrfs_block_group_cache *cache;
4596 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4599 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4601 btrfs_put_block_group(cache);
4606 * this function must be called within transaction
4608 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4609 struct btrfs_root *root,
4610 u64 bytenr, u64 num_bytes)
4612 struct btrfs_block_group_cache *cache;
4614 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4618 * pull in the free space cache (if any) so that our pin
4619 * removes the free space from the cache. We have load_only set
4620 * to one because the slow code to read in the free extents does check
4621 * the pinned extents.
4623 cache_block_group(cache, trans, root, 1);
4625 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4627 /* remove us from the free space cache (if we're there at all) */
4628 btrfs_remove_free_space(cache, bytenr, num_bytes);
4629 btrfs_put_block_group(cache);
4634 * btrfs_update_reserved_bytes - update the block_group and space info counters
4635 * @cache: The cache we are manipulating
4636 * @num_bytes: The number of bytes in question
4637 * @reserve: One of the reservation enums
4639 * This is called by the allocator when it reserves space, or by somebody who is
4640 * freeing space that was never actually used on disk. For example if you
4641 * reserve some space for a new leaf in transaction A and before transaction A
4642 * commits you free that leaf, you call this with reserve set to 0 in order to
4643 * clear the reservation.
4645 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4646 * ENOSPC accounting. For data we handle the reservation through clearing the
4647 * delalloc bits in the io_tree. We have to do this since we could end up
4648 * allocating less disk space for the amount of data we have reserved in the
4649 * case of compression.
4651 * If this is a reservation and the block group has become read only we cannot
4652 * make the reservation and return -EAGAIN, otherwise this function always
4655 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4656 u64 num_bytes, int reserve)
4658 struct btrfs_space_info *space_info = cache->space_info;
4660 spin_lock(&space_info->lock);
4661 spin_lock(&cache->lock);
4662 if (reserve != RESERVE_FREE) {
4666 cache->reserved += num_bytes;
4667 space_info->bytes_reserved += num_bytes;
4668 if (reserve == RESERVE_ALLOC) {
4669 BUG_ON(space_info->bytes_may_use < num_bytes);
4670 space_info->bytes_may_use -= num_bytes;
4675 space_info->bytes_readonly += num_bytes;
4676 cache->reserved -= num_bytes;
4677 space_info->bytes_reserved -= num_bytes;
4678 space_info->reservation_progress++;
4680 spin_unlock(&cache->lock);
4681 spin_unlock(&space_info->lock);
4685 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4686 struct btrfs_root *root)
4688 struct btrfs_fs_info *fs_info = root->fs_info;
4689 struct btrfs_caching_control *next;
4690 struct btrfs_caching_control *caching_ctl;
4691 struct btrfs_block_group_cache *cache;
4693 down_write(&fs_info->extent_commit_sem);
4695 list_for_each_entry_safe(caching_ctl, next,
4696 &fs_info->caching_block_groups, list) {
4697 cache = caching_ctl->block_group;
4698 if (block_group_cache_done(cache)) {
4699 cache->last_byte_to_unpin = (u64)-1;
4700 list_del_init(&caching_ctl->list);
4701 put_caching_control(caching_ctl);
4703 cache->last_byte_to_unpin = caching_ctl->progress;
4707 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4708 fs_info->pinned_extents = &fs_info->freed_extents[1];
4710 fs_info->pinned_extents = &fs_info->freed_extents[0];
4712 up_write(&fs_info->extent_commit_sem);
4714 update_global_block_rsv(fs_info);
4718 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4720 struct btrfs_fs_info *fs_info = root->fs_info;
4721 struct btrfs_block_group_cache *cache = NULL;
4724 while (start <= end) {
4726 start >= cache->key.objectid + cache->key.offset) {
4728 btrfs_put_block_group(cache);
4729 cache = btrfs_lookup_block_group(fs_info, start);
4733 len = cache->key.objectid + cache->key.offset - start;
4734 len = min(len, end + 1 - start);
4736 if (start < cache->last_byte_to_unpin) {
4737 len = min(len, cache->last_byte_to_unpin - start);
4738 btrfs_add_free_space(cache, start, len);
4743 spin_lock(&cache->space_info->lock);
4744 spin_lock(&cache->lock);
4745 cache->pinned -= len;
4746 cache->space_info->bytes_pinned -= len;
4748 cache->space_info->bytes_readonly += len;
4749 spin_unlock(&cache->lock);
4750 spin_unlock(&cache->space_info->lock);
4754 btrfs_put_block_group(cache);
4758 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4759 struct btrfs_root *root)
4761 struct btrfs_fs_info *fs_info = root->fs_info;
4762 struct extent_io_tree *unpin;
4767 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4768 unpin = &fs_info->freed_extents[1];
4770 unpin = &fs_info->freed_extents[0];
4773 ret = find_first_extent_bit(unpin, 0, &start, &end,
4778 if (btrfs_test_opt(root, DISCARD))
4779 ret = btrfs_discard_extent(root, start,
4780 end + 1 - start, NULL);
4782 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4783 unpin_extent_range(root, start, end);
4790 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4791 struct btrfs_root *root,
4792 u64 bytenr, u64 num_bytes, u64 parent,
4793 u64 root_objectid, u64 owner_objectid,
4794 u64 owner_offset, int refs_to_drop,
4795 struct btrfs_delayed_extent_op *extent_op)
4797 struct btrfs_key key;
4798 struct btrfs_path *path;
4799 struct btrfs_fs_info *info = root->fs_info;
4800 struct btrfs_root *extent_root = info->extent_root;
4801 struct extent_buffer *leaf;
4802 struct btrfs_extent_item *ei;
4803 struct btrfs_extent_inline_ref *iref;
4806 int extent_slot = 0;
4807 int found_extent = 0;
4812 path = btrfs_alloc_path();
4817 path->leave_spinning = 1;
4819 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4820 BUG_ON(!is_data && refs_to_drop != 1);
4822 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4823 bytenr, num_bytes, parent,
4824 root_objectid, owner_objectid,
4827 extent_slot = path->slots[0];
4828 while (extent_slot >= 0) {
4829 btrfs_item_key_to_cpu(path->nodes[0], &key,
4831 if (key.objectid != bytenr)
4833 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4834 key.offset == num_bytes) {
4838 if (path->slots[0] - extent_slot > 5)
4842 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4843 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4844 if (found_extent && item_size < sizeof(*ei))
4847 if (!found_extent) {
4849 ret = remove_extent_backref(trans, extent_root, path,
4853 btrfs_release_path(path);
4854 path->leave_spinning = 1;
4856 key.objectid = bytenr;
4857 key.type = BTRFS_EXTENT_ITEM_KEY;
4858 key.offset = num_bytes;
4860 ret = btrfs_search_slot(trans, extent_root,
4863 printk(KERN_ERR "umm, got %d back from search"
4864 ", was looking for %llu\n", ret,
4865 (unsigned long long)bytenr);
4867 btrfs_print_leaf(extent_root,
4871 extent_slot = path->slots[0];
4874 btrfs_print_leaf(extent_root, path->nodes[0]);
4876 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4877 "parent %llu root %llu owner %llu offset %llu\n",
4878 (unsigned long long)bytenr,
4879 (unsigned long long)parent,
4880 (unsigned long long)root_objectid,
4881 (unsigned long long)owner_objectid,
4882 (unsigned long long)owner_offset);
4885 leaf = path->nodes[0];
4886 item_size = btrfs_item_size_nr(leaf, extent_slot);
4887 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4888 if (item_size < sizeof(*ei)) {
4889 BUG_ON(found_extent || extent_slot != path->slots[0]);
4890 ret = convert_extent_item_v0(trans, extent_root, path,
4894 btrfs_release_path(path);
4895 path->leave_spinning = 1;
4897 key.objectid = bytenr;
4898 key.type = BTRFS_EXTENT_ITEM_KEY;
4899 key.offset = num_bytes;
4901 ret = btrfs_search_slot(trans, extent_root, &key, path,
4904 printk(KERN_ERR "umm, got %d back from search"
4905 ", was looking for %llu\n", ret,
4906 (unsigned long long)bytenr);
4907 btrfs_print_leaf(extent_root, path->nodes[0]);
4910 extent_slot = path->slots[0];
4911 leaf = path->nodes[0];
4912 item_size = btrfs_item_size_nr(leaf, extent_slot);
4915 BUG_ON(item_size < sizeof(*ei));
4916 ei = btrfs_item_ptr(leaf, extent_slot,
4917 struct btrfs_extent_item);
4918 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4919 struct btrfs_tree_block_info *bi;
4920 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4921 bi = (struct btrfs_tree_block_info *)(ei + 1);
4922 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4925 refs = btrfs_extent_refs(leaf, ei);
4926 BUG_ON(refs < refs_to_drop);
4927 refs -= refs_to_drop;
4931 __run_delayed_extent_op(extent_op, leaf, ei);
4933 * In the case of inline back ref, reference count will
4934 * be updated by remove_extent_backref
4937 BUG_ON(!found_extent);
4939 btrfs_set_extent_refs(leaf, ei, refs);
4940 btrfs_mark_buffer_dirty(leaf);
4943 ret = remove_extent_backref(trans, extent_root, path,
4950 BUG_ON(is_data && refs_to_drop !=
4951 extent_data_ref_count(root, path, iref));
4953 BUG_ON(path->slots[0] != extent_slot);
4955 BUG_ON(path->slots[0] != extent_slot + 1);
4956 path->slots[0] = extent_slot;
4961 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4964 btrfs_release_path(path);
4967 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4970 invalidate_mapping_pages(info->btree_inode->i_mapping,
4971 bytenr >> PAGE_CACHE_SHIFT,
4972 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4975 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4978 btrfs_free_path(path);
4983 * when we free an block, it is possible (and likely) that we free the last
4984 * delayed ref for that extent as well. This searches the delayed ref tree for
4985 * a given extent, and if there are no other delayed refs to be processed, it
4986 * removes it from the tree.
4988 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4989 struct btrfs_root *root, u64 bytenr)
4991 struct btrfs_delayed_ref_head *head;
4992 struct btrfs_delayed_ref_root *delayed_refs;
4993 struct btrfs_delayed_ref_node *ref;
4994 struct rb_node *node;
4997 delayed_refs = &trans->transaction->delayed_refs;
4998 spin_lock(&delayed_refs->lock);
4999 head = btrfs_find_delayed_ref_head(trans, bytenr);
5003 node = rb_prev(&head->node.rb_node);
5007 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5009 /* there are still entries for this ref, we can't drop it */
5010 if (ref->bytenr == bytenr)
5013 if (head->extent_op) {
5014 if (!head->must_insert_reserved)
5016 kfree(head->extent_op);
5017 head->extent_op = NULL;
5021 * waiting for the lock here would deadlock. If someone else has it
5022 * locked they are already in the process of dropping it anyway
5024 if (!mutex_trylock(&head->mutex))
5028 * at this point we have a head with no other entries. Go
5029 * ahead and process it.
5031 head->node.in_tree = 0;
5032 rb_erase(&head->node.rb_node, &delayed_refs->root);
5034 delayed_refs->num_entries--;
5035 if (waitqueue_active(&delayed_refs->seq_wait))
5036 wake_up(&delayed_refs->seq_wait);
5039 * we don't take a ref on the node because we're removing it from the
5040 * tree, so we just steal the ref the tree was holding.
5042 delayed_refs->num_heads--;
5043 if (list_empty(&head->cluster))
5044 delayed_refs->num_heads_ready--;
5046 list_del_init(&head->cluster);
5047 spin_unlock(&delayed_refs->lock);
5049 BUG_ON(head->extent_op);
5050 if (head->must_insert_reserved)
5053 mutex_unlock(&head->mutex);
5054 btrfs_put_delayed_ref(&head->node);
5057 spin_unlock(&delayed_refs->lock);
5061 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5062 struct btrfs_root *root,
5063 struct extent_buffer *buf,
5064 u64 parent, int last_ref, int for_cow)
5066 struct btrfs_block_group_cache *cache = NULL;
5069 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5070 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5071 buf->start, buf->len,
5072 parent, root->root_key.objectid,
5073 btrfs_header_level(buf),
5074 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5081 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5083 if (btrfs_header_generation(buf) == trans->transid) {
5084 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5085 ret = check_ref_cleanup(trans, root, buf->start);
5090 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5091 pin_down_extent(root, cache, buf->start, buf->len, 1);
5095 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5097 btrfs_add_free_space(cache, buf->start, buf->len);
5098 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5102 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5105 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5106 btrfs_put_block_group(cache);
5109 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5110 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5111 u64 owner, u64 offset, int for_cow)
5114 struct btrfs_fs_info *fs_info = root->fs_info;
5117 * tree log blocks never actually go into the extent allocation
5118 * tree, just update pinning info and exit early.
5120 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5121 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5122 /* unlocks the pinned mutex */
5123 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5125 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5126 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5128 parent, root_objectid, (int)owner,
5129 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5132 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5134 parent, root_objectid, owner,
5135 offset, BTRFS_DROP_DELAYED_REF,
5142 static u64 stripe_align(struct btrfs_root *root, u64 val)
5144 u64 mask = ((u64)root->stripesize - 1);
5145 u64 ret = (val + mask) & ~mask;
5150 * when we wait for progress in the block group caching, its because
5151 * our allocation attempt failed at least once. So, we must sleep
5152 * and let some progress happen before we try again.
5154 * This function will sleep at least once waiting for new free space to
5155 * show up, and then it will check the block group free space numbers
5156 * for our min num_bytes. Another option is to have it go ahead
5157 * and look in the rbtree for a free extent of a given size, but this
5161 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5164 struct btrfs_caching_control *caching_ctl;
5167 caching_ctl = get_caching_control(cache);
5171 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5172 (cache->free_space_ctl->free_space >= num_bytes));
5174 put_caching_control(caching_ctl);
5179 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5181 struct btrfs_caching_control *caching_ctl;
5184 caching_ctl = get_caching_control(cache);
5188 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5190 put_caching_control(caching_ctl);
5194 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5197 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5199 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5201 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5203 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5210 enum btrfs_loop_type {
5211 LOOP_FIND_IDEAL = 0,
5212 LOOP_CACHING_NOWAIT = 1,
5213 LOOP_CACHING_WAIT = 2,
5214 LOOP_ALLOC_CHUNK = 3,
5215 LOOP_NO_EMPTY_SIZE = 4,
5219 * walks the btree of allocated extents and find a hole of a given size.
5220 * The key ins is changed to record the hole:
5221 * ins->objectid == block start
5222 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5223 * ins->offset == number of blocks
5224 * Any available blocks before search_start are skipped.
5226 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5227 struct btrfs_root *orig_root,
5228 u64 num_bytes, u64 empty_size,
5229 u64 search_start, u64 search_end,
5230 u64 hint_byte, struct btrfs_key *ins,
5234 struct btrfs_root *root = orig_root->fs_info->extent_root;
5235 struct btrfs_free_cluster *last_ptr = NULL;
5236 struct btrfs_block_group_cache *block_group = NULL;
5237 struct btrfs_block_group_cache *used_block_group;
5238 int empty_cluster = 2 * 1024 * 1024;
5239 int allowed_chunk_alloc = 0;
5240 int done_chunk_alloc = 0;
5241 struct btrfs_space_info *space_info;
5244 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5245 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5246 bool found_uncached_bg = false;
5247 bool failed_cluster_refill = false;
5248 bool failed_alloc = false;
5249 bool use_cluster = true;
5250 bool have_caching_bg = false;
5251 u64 ideal_cache_percent = 0;
5252 u64 ideal_cache_offset = 0;
5254 WARN_ON(num_bytes < root->sectorsize);
5255 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5259 space_info = __find_space_info(root->fs_info, data);
5261 printk(KERN_ERR "No space info for %llu\n", data);
5266 * If the space info is for both data and metadata it means we have a
5267 * small filesystem and we can't use the clustering stuff.
5269 if (btrfs_mixed_space_info(space_info))
5270 use_cluster = false;
5272 if (orig_root->ref_cows || empty_size)
5273 allowed_chunk_alloc = 1;
5275 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5276 last_ptr = &root->fs_info->meta_alloc_cluster;
5277 if (!btrfs_test_opt(root, SSD))
5278 empty_cluster = 64 * 1024;
5281 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5282 btrfs_test_opt(root, SSD)) {
5283 last_ptr = &root->fs_info->data_alloc_cluster;
5287 spin_lock(&last_ptr->lock);
5288 if (last_ptr->block_group)
5289 hint_byte = last_ptr->window_start;
5290 spin_unlock(&last_ptr->lock);
5293 search_start = max(search_start, first_logical_byte(root, 0));
5294 search_start = max(search_start, hint_byte);
5299 if (search_start == hint_byte) {
5301 block_group = btrfs_lookup_block_group(root->fs_info,
5303 used_block_group = block_group;
5305 * we don't want to use the block group if it doesn't match our
5306 * allocation bits, or if its not cached.
5308 * However if we are re-searching with an ideal block group
5309 * picked out then we don't care that the block group is cached.
5311 if (block_group && block_group_bits(block_group, data) &&
5312 (block_group->cached != BTRFS_CACHE_NO ||
5313 search_start == ideal_cache_offset)) {
5314 down_read(&space_info->groups_sem);
5315 if (list_empty(&block_group->list) ||
5318 * someone is removing this block group,
5319 * we can't jump into the have_block_group
5320 * target because our list pointers are not
5323 btrfs_put_block_group(block_group);
5324 up_read(&space_info->groups_sem);
5326 index = get_block_group_index(block_group);
5327 goto have_block_group;
5329 } else if (block_group) {
5330 btrfs_put_block_group(block_group);
5334 have_caching_bg = false;
5335 down_read(&space_info->groups_sem);
5336 list_for_each_entry(block_group, &space_info->block_groups[index],
5341 used_block_group = block_group;
5342 btrfs_get_block_group(block_group);
5343 search_start = block_group->key.objectid;
5346 * this can happen if we end up cycling through all the
5347 * raid types, but we want to make sure we only allocate
5348 * for the proper type.
5350 if (!block_group_bits(block_group, data)) {
5351 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5352 BTRFS_BLOCK_GROUP_RAID1 |
5353 BTRFS_BLOCK_GROUP_RAID10;
5356 * if they asked for extra copies and this block group
5357 * doesn't provide them, bail. This does allow us to
5358 * fill raid0 from raid1.
5360 if ((data & extra) && !(block_group->flags & extra))
5365 cached = block_group_cache_done(block_group);
5366 if (unlikely(!cached)) {
5369 found_uncached_bg = true;
5370 ret = cache_block_group(block_group, trans,
5372 if (block_group->cached == BTRFS_CACHE_FINISHED)
5375 free_percent = btrfs_block_group_used(&block_group->item);
5376 free_percent *= 100;
5377 free_percent = div64_u64(free_percent,
5378 block_group->key.offset);
5379 free_percent = 100 - free_percent;
5380 if (free_percent > ideal_cache_percent &&
5381 likely(!block_group->ro)) {
5382 ideal_cache_offset = block_group->key.objectid;
5383 ideal_cache_percent = free_percent;
5387 * The caching workers are limited to 2 threads, so we
5388 * can queue as much work as we care to.
5390 if (loop > LOOP_FIND_IDEAL) {
5391 ret = cache_block_group(block_group, trans,
5397 * If loop is set for cached only, try the next block
5400 if (loop == LOOP_FIND_IDEAL)
5405 if (unlikely(block_group->ro))
5409 * Ok we want to try and use the cluster allocator, so
5414 * the refill lock keeps out other
5415 * people trying to start a new cluster
5417 spin_lock(&last_ptr->refill_lock);
5418 used_block_group = last_ptr->block_group;
5419 if (used_block_group != block_group &&
5420 (!used_block_group ||
5421 used_block_group->ro ||
5422 !block_group_bits(used_block_group, data))) {
5423 used_block_group = block_group;
5424 goto refill_cluster;
5427 if (used_block_group != block_group)
5428 btrfs_get_block_group(used_block_group);
5430 offset = btrfs_alloc_from_cluster(used_block_group,
5431 last_ptr, num_bytes, used_block_group->key.objectid);
5433 /* we have a block, we're done */
5434 spin_unlock(&last_ptr->refill_lock);
5438 WARN_ON(last_ptr->block_group != used_block_group);
5439 if (used_block_group != block_group) {
5440 btrfs_put_block_group(used_block_group);
5441 used_block_group = block_group;
5444 BUG_ON(used_block_group != block_group);
5445 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5446 * set up a new clusters, so lets just skip it
5447 * and let the allocator find whatever block
5448 * it can find. If we reach this point, we
5449 * will have tried the cluster allocator
5450 * plenty of times and not have found
5451 * anything, so we are likely way too
5452 * fragmented for the clustering stuff to find
5455 * However, if the cluster is taken from the
5456 * current block group, release the cluster
5457 * first, so that we stand a better chance of
5458 * succeeding in the unclustered
5460 if (loop >= LOOP_NO_EMPTY_SIZE &&
5461 last_ptr->block_group != block_group) {
5462 spin_unlock(&last_ptr->refill_lock);
5463 goto unclustered_alloc;
5467 * this cluster didn't work out, free it and
5470 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5472 if (loop >= LOOP_NO_EMPTY_SIZE) {
5473 spin_unlock(&last_ptr->refill_lock);
5474 goto unclustered_alloc;
5477 /* allocate a cluster in this block group */
5478 ret = btrfs_find_space_cluster(trans, root,
5479 block_group, last_ptr,
5480 search_start, num_bytes,
5481 empty_cluster + empty_size);
5484 * now pull our allocation out of this
5487 offset = btrfs_alloc_from_cluster(block_group,
5488 last_ptr, num_bytes,
5491 /* we found one, proceed */
5492 spin_unlock(&last_ptr->refill_lock);
5495 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5496 && !failed_cluster_refill) {
5497 spin_unlock(&last_ptr->refill_lock);
5499 failed_cluster_refill = true;
5500 wait_block_group_cache_progress(block_group,
5501 num_bytes + empty_cluster + empty_size);
5502 goto have_block_group;
5506 * at this point we either didn't find a cluster
5507 * or we weren't able to allocate a block from our
5508 * cluster. Free the cluster we've been trying
5509 * to use, and go to the next block group
5511 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5512 spin_unlock(&last_ptr->refill_lock);
5517 spin_lock(&block_group->free_space_ctl->tree_lock);
5519 block_group->free_space_ctl->free_space <
5520 num_bytes + empty_cluster + empty_size) {
5521 spin_unlock(&block_group->free_space_ctl->tree_lock);
5524 spin_unlock(&block_group->free_space_ctl->tree_lock);
5526 offset = btrfs_find_space_for_alloc(block_group, search_start,
5527 num_bytes, empty_size);
5529 * If we didn't find a chunk, and we haven't failed on this
5530 * block group before, and this block group is in the middle of
5531 * caching and we are ok with waiting, then go ahead and wait
5532 * for progress to be made, and set failed_alloc to true.
5534 * If failed_alloc is true then we've already waited on this
5535 * block group once and should move on to the next block group.
5537 if (!offset && !failed_alloc && !cached &&
5538 loop > LOOP_CACHING_NOWAIT) {
5539 wait_block_group_cache_progress(block_group,
5540 num_bytes + empty_size);
5541 failed_alloc = true;
5542 goto have_block_group;
5543 } else if (!offset) {
5545 have_caching_bg = true;
5549 search_start = stripe_align(root, offset);
5550 /* move on to the next group */
5551 if (search_start + num_bytes >= search_end) {
5552 btrfs_add_free_space(used_block_group, offset, num_bytes);
5556 /* move on to the next group */
5557 if (search_start + num_bytes >
5558 used_block_group->key.objectid + used_block_group->key.offset) {
5559 btrfs_add_free_space(used_block_group, offset, num_bytes);
5563 if (offset < search_start)
5564 btrfs_add_free_space(used_block_group, offset,
5565 search_start - offset);
5566 BUG_ON(offset > search_start);
5568 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5570 if (ret == -EAGAIN) {
5571 btrfs_add_free_space(used_block_group, offset, num_bytes);
5575 /* we are all good, lets return */
5576 ins->objectid = search_start;
5577 ins->offset = num_bytes;
5579 if (offset < search_start)
5580 btrfs_add_free_space(used_block_group, offset,
5581 search_start - offset);
5582 BUG_ON(offset > search_start);
5583 if (used_block_group != block_group)
5584 btrfs_put_block_group(used_block_group);
5585 btrfs_put_block_group(block_group);
5588 failed_cluster_refill = false;
5589 failed_alloc = false;
5590 BUG_ON(index != get_block_group_index(block_group));
5591 if (used_block_group != block_group)
5592 btrfs_put_block_group(used_block_group);
5593 btrfs_put_block_group(block_group);
5595 up_read(&space_info->groups_sem);
5597 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5600 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5603 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5604 * for them to make caching progress. Also
5605 * determine the best possible bg to cache
5606 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5607 * caching kthreads as we move along
5608 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5609 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5610 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5613 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5615 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5616 found_uncached_bg = false;
5618 if (!ideal_cache_percent)
5622 * 1 of the following 2 things have happened so far
5624 * 1) We found an ideal block group for caching that
5625 * is mostly full and will cache quickly, so we might
5626 * as well wait for it.
5628 * 2) We searched for cached only and we didn't find
5629 * anything, and we didn't start any caching kthreads
5630 * either, so chances are we will loop through and
5631 * start a couple caching kthreads, and then come back
5632 * around and just wait for them. This will be slower
5633 * because we will have 2 caching kthreads reading at
5634 * the same time when we could have just started one
5635 * and waited for it to get far enough to give us an
5636 * allocation, so go ahead and go to the wait caching
5639 loop = LOOP_CACHING_WAIT;
5640 search_start = ideal_cache_offset;
5641 ideal_cache_percent = 0;
5643 } else if (loop == LOOP_FIND_IDEAL) {
5645 * Didn't find a uncached bg, wait on anything we find
5648 loop = LOOP_CACHING_WAIT;
5654 if (loop == LOOP_ALLOC_CHUNK) {
5655 if (allowed_chunk_alloc) {
5656 ret = do_chunk_alloc(trans, root, num_bytes +
5657 2 * 1024 * 1024, data,
5658 CHUNK_ALLOC_LIMITED);
5659 allowed_chunk_alloc = 0;
5661 done_chunk_alloc = 1;
5662 } else if (!done_chunk_alloc &&
5663 space_info->force_alloc ==
5664 CHUNK_ALLOC_NO_FORCE) {
5665 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5669 * We didn't allocate a chunk, go ahead and drop the
5670 * empty size and loop again.
5672 if (!done_chunk_alloc)
5673 loop = LOOP_NO_EMPTY_SIZE;
5676 if (loop == LOOP_NO_EMPTY_SIZE) {
5682 } else if (!ins->objectid) {
5684 } else if (ins->objectid) {
5691 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5692 int dump_block_groups)
5694 struct btrfs_block_group_cache *cache;
5697 spin_lock(&info->lock);
5698 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5699 (unsigned long long)info->flags,
5700 (unsigned long long)(info->total_bytes - info->bytes_used -
5701 info->bytes_pinned - info->bytes_reserved -
5702 info->bytes_readonly),
5703 (info->full) ? "" : "not ");
5704 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5705 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5706 (unsigned long long)info->total_bytes,
5707 (unsigned long long)info->bytes_used,
5708 (unsigned long long)info->bytes_pinned,
5709 (unsigned long long)info->bytes_reserved,
5710 (unsigned long long)info->bytes_may_use,
5711 (unsigned long long)info->bytes_readonly);
5712 spin_unlock(&info->lock);
5714 if (!dump_block_groups)
5717 down_read(&info->groups_sem);
5719 list_for_each_entry(cache, &info->block_groups[index], list) {
5720 spin_lock(&cache->lock);
5721 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5722 "%llu pinned %llu reserved\n",
5723 (unsigned long long)cache->key.objectid,
5724 (unsigned long long)cache->key.offset,
5725 (unsigned long long)btrfs_block_group_used(&cache->item),
5726 (unsigned long long)cache->pinned,
5727 (unsigned long long)cache->reserved);
5728 btrfs_dump_free_space(cache, bytes);
5729 spin_unlock(&cache->lock);
5731 if (++index < BTRFS_NR_RAID_TYPES)
5733 up_read(&info->groups_sem);
5736 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5737 struct btrfs_root *root,
5738 u64 num_bytes, u64 min_alloc_size,
5739 u64 empty_size, u64 hint_byte,
5740 u64 search_end, struct btrfs_key *ins,
5744 u64 search_start = 0;
5746 data = btrfs_get_alloc_profile(root, data);
5749 * the only place that sets empty_size is btrfs_realloc_node, which
5750 * is not called recursively on allocations
5752 if (empty_size || root->ref_cows)
5753 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5754 num_bytes + 2 * 1024 * 1024, data,
5755 CHUNK_ALLOC_NO_FORCE);
5757 WARN_ON(num_bytes < root->sectorsize);
5758 ret = find_free_extent(trans, root, num_bytes, empty_size,
5759 search_start, search_end, hint_byte,
5762 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5763 num_bytes = num_bytes >> 1;
5764 num_bytes = num_bytes & ~(root->sectorsize - 1);
5765 num_bytes = max(num_bytes, min_alloc_size);
5766 do_chunk_alloc(trans, root->fs_info->extent_root,
5767 num_bytes, data, CHUNK_ALLOC_FORCE);
5770 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5771 struct btrfs_space_info *sinfo;
5773 sinfo = __find_space_info(root->fs_info, data);
5774 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5775 "wanted %llu\n", (unsigned long long)data,
5776 (unsigned long long)num_bytes);
5777 dump_space_info(sinfo, num_bytes, 1);
5780 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5785 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5786 u64 start, u64 len, int pin)
5788 struct btrfs_block_group_cache *cache;
5791 cache = btrfs_lookup_block_group(root->fs_info, start);
5793 printk(KERN_ERR "Unable to find block group for %llu\n",
5794 (unsigned long long)start);
5798 if (btrfs_test_opt(root, DISCARD))
5799 ret = btrfs_discard_extent(root, start, len, NULL);
5802 pin_down_extent(root, cache, start, len, 1);
5804 btrfs_add_free_space(cache, start, len);
5805 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5807 btrfs_put_block_group(cache);
5809 trace_btrfs_reserved_extent_free(root, start, len);
5814 int btrfs_free_reserved_extent(struct btrfs_root *root,
5817 return __btrfs_free_reserved_extent(root, start, len, 0);
5820 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5823 return __btrfs_free_reserved_extent(root, start, len, 1);
5826 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5827 struct btrfs_root *root,
5828 u64 parent, u64 root_objectid,
5829 u64 flags, u64 owner, u64 offset,
5830 struct btrfs_key *ins, int ref_mod)
5833 struct btrfs_fs_info *fs_info = root->fs_info;
5834 struct btrfs_extent_item *extent_item;
5835 struct btrfs_extent_inline_ref *iref;
5836 struct btrfs_path *path;
5837 struct extent_buffer *leaf;
5842 type = BTRFS_SHARED_DATA_REF_KEY;
5844 type = BTRFS_EXTENT_DATA_REF_KEY;
5846 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5848 path = btrfs_alloc_path();
5852 path->leave_spinning = 1;
5853 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5857 leaf = path->nodes[0];
5858 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5859 struct btrfs_extent_item);
5860 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5861 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5862 btrfs_set_extent_flags(leaf, extent_item,
5863 flags | BTRFS_EXTENT_FLAG_DATA);
5865 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5866 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5868 struct btrfs_shared_data_ref *ref;
5869 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5870 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5871 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5873 struct btrfs_extent_data_ref *ref;
5874 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5875 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5876 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5877 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5878 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5881 btrfs_mark_buffer_dirty(path->nodes[0]);
5882 btrfs_free_path(path);
5884 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5886 printk(KERN_ERR "btrfs update block group failed for %llu "
5887 "%llu\n", (unsigned long long)ins->objectid,
5888 (unsigned long long)ins->offset);
5894 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5895 struct btrfs_root *root,
5896 u64 parent, u64 root_objectid,
5897 u64 flags, struct btrfs_disk_key *key,
5898 int level, struct btrfs_key *ins)
5901 struct btrfs_fs_info *fs_info = root->fs_info;
5902 struct btrfs_extent_item *extent_item;
5903 struct btrfs_tree_block_info *block_info;
5904 struct btrfs_extent_inline_ref *iref;
5905 struct btrfs_path *path;
5906 struct extent_buffer *leaf;
5907 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5909 path = btrfs_alloc_path();
5913 path->leave_spinning = 1;
5914 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5918 leaf = path->nodes[0];
5919 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5920 struct btrfs_extent_item);
5921 btrfs_set_extent_refs(leaf, extent_item, 1);
5922 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5923 btrfs_set_extent_flags(leaf, extent_item,
5924 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5925 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5927 btrfs_set_tree_block_key(leaf, block_info, key);
5928 btrfs_set_tree_block_level(leaf, block_info, level);
5930 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5932 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5933 btrfs_set_extent_inline_ref_type(leaf, iref,
5934 BTRFS_SHARED_BLOCK_REF_KEY);
5935 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5937 btrfs_set_extent_inline_ref_type(leaf, iref,
5938 BTRFS_TREE_BLOCK_REF_KEY);
5939 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5942 btrfs_mark_buffer_dirty(leaf);
5943 btrfs_free_path(path);
5945 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5947 printk(KERN_ERR "btrfs update block group failed for %llu "
5948 "%llu\n", (unsigned long long)ins->objectid,
5949 (unsigned long long)ins->offset);
5955 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5956 struct btrfs_root *root,
5957 u64 root_objectid, u64 owner,
5958 u64 offset, struct btrfs_key *ins)
5962 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5964 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
5966 root_objectid, owner, offset,
5967 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
5972 * this is used by the tree logging recovery code. It records that
5973 * an extent has been allocated and makes sure to clear the free
5974 * space cache bits as well
5976 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5977 struct btrfs_root *root,
5978 u64 root_objectid, u64 owner, u64 offset,
5979 struct btrfs_key *ins)
5982 struct btrfs_block_group_cache *block_group;
5983 struct btrfs_caching_control *caching_ctl;
5984 u64 start = ins->objectid;
5985 u64 num_bytes = ins->offset;
5987 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5988 cache_block_group(block_group, trans, NULL, 0);
5989 caching_ctl = get_caching_control(block_group);
5992 BUG_ON(!block_group_cache_done(block_group));
5993 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5996 mutex_lock(&caching_ctl->mutex);
5998 if (start >= caching_ctl->progress) {
5999 ret = add_excluded_extent(root, start, num_bytes);
6001 } else if (start + num_bytes <= caching_ctl->progress) {
6002 ret = btrfs_remove_free_space(block_group,
6006 num_bytes = caching_ctl->progress - start;
6007 ret = btrfs_remove_free_space(block_group,
6011 start = caching_ctl->progress;
6012 num_bytes = ins->objectid + ins->offset -
6013 caching_ctl->progress;
6014 ret = add_excluded_extent(root, start, num_bytes);
6018 mutex_unlock(&caching_ctl->mutex);
6019 put_caching_control(caching_ctl);
6022 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6023 RESERVE_ALLOC_NO_ACCOUNT);
6025 btrfs_put_block_group(block_group);
6026 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6027 0, owner, offset, ins, 1);
6031 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6032 struct btrfs_root *root,
6033 u64 bytenr, u32 blocksize,
6036 struct extent_buffer *buf;
6038 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6040 return ERR_PTR(-ENOMEM);
6041 btrfs_set_header_generation(buf, trans->transid);
6042 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6043 btrfs_tree_lock(buf);
6044 clean_tree_block(trans, root, buf);
6046 btrfs_set_lock_blocking(buf);
6047 btrfs_set_buffer_uptodate(buf);
6049 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6051 * we allow two log transactions at a time, use different
6052 * EXENT bit to differentiate dirty pages.
6054 if (root->log_transid % 2 == 0)
6055 set_extent_dirty(&root->dirty_log_pages, buf->start,
6056 buf->start + buf->len - 1, GFP_NOFS);
6058 set_extent_new(&root->dirty_log_pages, buf->start,
6059 buf->start + buf->len - 1, GFP_NOFS);
6061 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6062 buf->start + buf->len - 1, GFP_NOFS);
6064 trans->blocks_used++;
6065 /* this returns a buffer locked for blocking */
6069 static struct btrfs_block_rsv *
6070 use_block_rsv(struct btrfs_trans_handle *trans,
6071 struct btrfs_root *root, u32 blocksize)
6073 struct btrfs_block_rsv *block_rsv;
6074 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6077 block_rsv = get_block_rsv(trans, root);
6079 if (block_rsv->size == 0) {
6080 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6082 * If we couldn't reserve metadata bytes try and use some from
6083 * the global reserve.
6085 if (ret && block_rsv != global_rsv) {
6086 ret = block_rsv_use_bytes(global_rsv, blocksize);
6089 return ERR_PTR(ret);
6091 return ERR_PTR(ret);
6096 ret = block_rsv_use_bytes(block_rsv, blocksize);
6100 static DEFINE_RATELIMIT_STATE(_rs,
6101 DEFAULT_RATELIMIT_INTERVAL,
6102 /*DEFAULT_RATELIMIT_BURST*/ 2);
6103 if (__ratelimit(&_rs)) {
6104 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6107 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6110 } else if (ret && block_rsv != global_rsv) {
6111 ret = block_rsv_use_bytes(global_rsv, blocksize);
6117 return ERR_PTR(-ENOSPC);
6120 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
6122 block_rsv_add_bytes(block_rsv, blocksize, 0);
6123 block_rsv_release_bytes(block_rsv, NULL, 0);
6127 * finds a free extent and does all the dirty work required for allocation
6128 * returns the key for the extent through ins, and a tree buffer for
6129 * the first block of the extent through buf.
6131 * returns the tree buffer or NULL.
6133 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6134 struct btrfs_root *root, u32 blocksize,
6135 u64 parent, u64 root_objectid,
6136 struct btrfs_disk_key *key, int level,
6137 u64 hint, u64 empty_size, int for_cow)
6139 struct btrfs_key ins;
6140 struct btrfs_block_rsv *block_rsv;
6141 struct extent_buffer *buf;
6146 block_rsv = use_block_rsv(trans, root, blocksize);
6147 if (IS_ERR(block_rsv))
6148 return ERR_CAST(block_rsv);
6150 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6151 empty_size, hint, (u64)-1, &ins, 0);
6153 unuse_block_rsv(block_rsv, blocksize);
6154 return ERR_PTR(ret);
6157 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6159 BUG_ON(IS_ERR(buf));
6161 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6163 parent = ins.objectid;
6164 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6168 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6169 struct btrfs_delayed_extent_op *extent_op;
6170 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6173 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6175 memset(&extent_op->key, 0, sizeof(extent_op->key));
6176 extent_op->flags_to_set = flags;
6177 extent_op->update_key = 1;
6178 extent_op->update_flags = 1;
6179 extent_op->is_data = 0;
6181 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6183 ins.offset, parent, root_objectid,
6184 level, BTRFS_ADD_DELAYED_EXTENT,
6185 extent_op, for_cow);
6191 struct walk_control {
6192 u64 refs[BTRFS_MAX_LEVEL];
6193 u64 flags[BTRFS_MAX_LEVEL];
6194 struct btrfs_key update_progress;
6205 #define DROP_REFERENCE 1
6206 #define UPDATE_BACKREF 2
6208 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6209 struct btrfs_root *root,
6210 struct walk_control *wc,
6211 struct btrfs_path *path)
6219 struct btrfs_key key;
6220 struct extent_buffer *eb;
6225 if (path->slots[wc->level] < wc->reada_slot) {
6226 wc->reada_count = wc->reada_count * 2 / 3;
6227 wc->reada_count = max(wc->reada_count, 2);
6229 wc->reada_count = wc->reada_count * 3 / 2;
6230 wc->reada_count = min_t(int, wc->reada_count,
6231 BTRFS_NODEPTRS_PER_BLOCK(root));
6234 eb = path->nodes[wc->level];
6235 nritems = btrfs_header_nritems(eb);
6236 blocksize = btrfs_level_size(root, wc->level - 1);
6238 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6239 if (nread >= wc->reada_count)
6243 bytenr = btrfs_node_blockptr(eb, slot);
6244 generation = btrfs_node_ptr_generation(eb, slot);
6246 if (slot == path->slots[wc->level])
6249 if (wc->stage == UPDATE_BACKREF &&
6250 generation <= root->root_key.offset)
6253 /* We don't lock the tree block, it's OK to be racy here */
6254 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6259 if (wc->stage == DROP_REFERENCE) {
6263 if (wc->level == 1 &&
6264 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6266 if (!wc->update_ref ||
6267 generation <= root->root_key.offset)
6269 btrfs_node_key_to_cpu(eb, &key, slot);
6270 ret = btrfs_comp_cpu_keys(&key,
6271 &wc->update_progress);
6275 if (wc->level == 1 &&
6276 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6280 ret = readahead_tree_block(root, bytenr, blocksize,
6286 wc->reada_slot = slot;
6290 * hepler to process tree block while walking down the tree.
6292 * when wc->stage == UPDATE_BACKREF, this function updates
6293 * back refs for pointers in the block.
6295 * NOTE: return value 1 means we should stop walking down.
6297 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6298 struct btrfs_root *root,
6299 struct btrfs_path *path,
6300 struct walk_control *wc, int lookup_info)
6302 int level = wc->level;
6303 struct extent_buffer *eb = path->nodes[level];
6304 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6307 if (wc->stage == UPDATE_BACKREF &&
6308 btrfs_header_owner(eb) != root->root_key.objectid)
6312 * when reference count of tree block is 1, it won't increase
6313 * again. once full backref flag is set, we never clear it.
6316 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6317 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6318 BUG_ON(!path->locks[level]);
6319 ret = btrfs_lookup_extent_info(trans, root,
6324 BUG_ON(wc->refs[level] == 0);
6327 if (wc->stage == DROP_REFERENCE) {
6328 if (wc->refs[level] > 1)
6331 if (path->locks[level] && !wc->keep_locks) {
6332 btrfs_tree_unlock_rw(eb, path->locks[level]);
6333 path->locks[level] = 0;
6338 /* wc->stage == UPDATE_BACKREF */
6339 if (!(wc->flags[level] & flag)) {
6340 BUG_ON(!path->locks[level]);
6341 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6343 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6345 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6348 wc->flags[level] |= flag;
6352 * the block is shared by multiple trees, so it's not good to
6353 * keep the tree lock
6355 if (path->locks[level] && level > 0) {
6356 btrfs_tree_unlock_rw(eb, path->locks[level]);
6357 path->locks[level] = 0;
6363 * hepler to process tree block pointer.
6365 * when wc->stage == DROP_REFERENCE, this function checks
6366 * reference count of the block pointed to. if the block
6367 * is shared and we need update back refs for the subtree
6368 * rooted at the block, this function changes wc->stage to
6369 * UPDATE_BACKREF. if the block is shared and there is no
6370 * need to update back, this function drops the reference
6373 * NOTE: return value 1 means we should stop walking down.
6375 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6376 struct btrfs_root *root,
6377 struct btrfs_path *path,
6378 struct walk_control *wc, int *lookup_info)
6384 struct btrfs_key key;
6385 struct extent_buffer *next;
6386 int level = wc->level;
6390 generation = btrfs_node_ptr_generation(path->nodes[level],
6391 path->slots[level]);
6393 * if the lower level block was created before the snapshot
6394 * was created, we know there is no need to update back refs
6397 if (wc->stage == UPDATE_BACKREF &&
6398 generation <= root->root_key.offset) {
6403 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6404 blocksize = btrfs_level_size(root, level - 1);
6406 next = btrfs_find_tree_block(root, bytenr, blocksize);
6408 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6413 btrfs_tree_lock(next);
6414 btrfs_set_lock_blocking(next);
6416 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6417 &wc->refs[level - 1],
6418 &wc->flags[level - 1]);
6420 BUG_ON(wc->refs[level - 1] == 0);
6423 if (wc->stage == DROP_REFERENCE) {
6424 if (wc->refs[level - 1] > 1) {
6426 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6429 if (!wc->update_ref ||
6430 generation <= root->root_key.offset)
6433 btrfs_node_key_to_cpu(path->nodes[level], &key,
6434 path->slots[level]);
6435 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6439 wc->stage = UPDATE_BACKREF;
6440 wc->shared_level = level - 1;
6444 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6448 if (!btrfs_buffer_uptodate(next, generation)) {
6449 btrfs_tree_unlock(next);
6450 free_extent_buffer(next);
6456 if (reada && level == 1)
6457 reada_walk_down(trans, root, wc, path);
6458 next = read_tree_block(root, bytenr, blocksize, generation);
6461 btrfs_tree_lock(next);
6462 btrfs_set_lock_blocking(next);
6466 BUG_ON(level != btrfs_header_level(next));
6467 path->nodes[level] = next;
6468 path->slots[level] = 0;
6469 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6475 wc->refs[level - 1] = 0;
6476 wc->flags[level - 1] = 0;
6477 if (wc->stage == DROP_REFERENCE) {
6478 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6479 parent = path->nodes[level]->start;
6481 BUG_ON(root->root_key.objectid !=
6482 btrfs_header_owner(path->nodes[level]));
6486 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6487 root->root_key.objectid, level - 1, 0, 0);
6490 btrfs_tree_unlock(next);
6491 free_extent_buffer(next);
6497 * hepler to process tree block while walking up the tree.
6499 * when wc->stage == DROP_REFERENCE, this function drops
6500 * reference count on the block.
6502 * when wc->stage == UPDATE_BACKREF, this function changes
6503 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6504 * to UPDATE_BACKREF previously while processing the block.
6506 * NOTE: return value 1 means we should stop walking up.
6508 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6509 struct btrfs_root *root,
6510 struct btrfs_path *path,
6511 struct walk_control *wc)
6514 int level = wc->level;
6515 struct extent_buffer *eb = path->nodes[level];
6518 if (wc->stage == UPDATE_BACKREF) {
6519 BUG_ON(wc->shared_level < level);
6520 if (level < wc->shared_level)
6523 ret = find_next_key(path, level + 1, &wc->update_progress);
6527 wc->stage = DROP_REFERENCE;
6528 wc->shared_level = -1;
6529 path->slots[level] = 0;
6532 * check reference count again if the block isn't locked.
6533 * we should start walking down the tree again if reference
6536 if (!path->locks[level]) {
6538 btrfs_tree_lock(eb);
6539 btrfs_set_lock_blocking(eb);
6540 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6542 ret = btrfs_lookup_extent_info(trans, root,
6547 BUG_ON(wc->refs[level] == 0);
6548 if (wc->refs[level] == 1) {
6549 btrfs_tree_unlock_rw(eb, path->locks[level]);
6555 /* wc->stage == DROP_REFERENCE */
6556 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6558 if (wc->refs[level] == 1) {
6560 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6561 ret = btrfs_dec_ref(trans, root, eb, 1,
6564 ret = btrfs_dec_ref(trans, root, eb, 0,
6568 /* make block locked assertion in clean_tree_block happy */
6569 if (!path->locks[level] &&
6570 btrfs_header_generation(eb) == trans->transid) {
6571 btrfs_tree_lock(eb);
6572 btrfs_set_lock_blocking(eb);
6573 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6575 clean_tree_block(trans, root, eb);
6578 if (eb == root->node) {
6579 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6582 BUG_ON(root->root_key.objectid !=
6583 btrfs_header_owner(eb));
6585 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6586 parent = path->nodes[level + 1]->start;
6588 BUG_ON(root->root_key.objectid !=
6589 btrfs_header_owner(path->nodes[level + 1]));
6592 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
6594 wc->refs[level] = 0;
6595 wc->flags[level] = 0;
6599 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6600 struct btrfs_root *root,
6601 struct btrfs_path *path,
6602 struct walk_control *wc)
6604 int level = wc->level;
6605 int lookup_info = 1;
6608 while (level >= 0) {
6609 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6616 if (path->slots[level] >=
6617 btrfs_header_nritems(path->nodes[level]))
6620 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6622 path->slots[level]++;
6631 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6632 struct btrfs_root *root,
6633 struct btrfs_path *path,
6634 struct walk_control *wc, int max_level)
6636 int level = wc->level;
6639 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6640 while (level < max_level && path->nodes[level]) {
6642 if (path->slots[level] + 1 <
6643 btrfs_header_nritems(path->nodes[level])) {
6644 path->slots[level]++;
6647 ret = walk_up_proc(trans, root, path, wc);
6651 if (path->locks[level]) {
6652 btrfs_tree_unlock_rw(path->nodes[level],
6653 path->locks[level]);
6654 path->locks[level] = 0;
6656 free_extent_buffer(path->nodes[level]);
6657 path->nodes[level] = NULL;
6665 * drop a subvolume tree.
6667 * this function traverses the tree freeing any blocks that only
6668 * referenced by the tree.
6670 * when a shared tree block is found. this function decreases its
6671 * reference count by one. if update_ref is true, this function
6672 * also make sure backrefs for the shared block and all lower level
6673 * blocks are properly updated.
6675 void btrfs_drop_snapshot(struct btrfs_root *root,
6676 struct btrfs_block_rsv *block_rsv, int update_ref,
6679 struct btrfs_path *path;
6680 struct btrfs_trans_handle *trans;
6681 struct btrfs_root *tree_root = root->fs_info->tree_root;
6682 struct btrfs_root_item *root_item = &root->root_item;
6683 struct walk_control *wc;
6684 struct btrfs_key key;
6689 path = btrfs_alloc_path();
6695 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6697 btrfs_free_path(path);
6702 trans = btrfs_start_transaction(tree_root, 0);
6703 BUG_ON(IS_ERR(trans));
6706 trans->block_rsv = block_rsv;
6708 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6709 level = btrfs_header_level(root->node);
6710 path->nodes[level] = btrfs_lock_root_node(root);
6711 btrfs_set_lock_blocking(path->nodes[level]);
6712 path->slots[level] = 0;
6713 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6714 memset(&wc->update_progress, 0,
6715 sizeof(wc->update_progress));
6717 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6718 memcpy(&wc->update_progress, &key,
6719 sizeof(wc->update_progress));
6721 level = root_item->drop_level;
6723 path->lowest_level = level;
6724 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6725 path->lowest_level = 0;
6733 * unlock our path, this is safe because only this
6734 * function is allowed to delete this snapshot
6736 btrfs_unlock_up_safe(path, 0);
6738 level = btrfs_header_level(root->node);
6740 btrfs_tree_lock(path->nodes[level]);
6741 btrfs_set_lock_blocking(path->nodes[level]);
6743 ret = btrfs_lookup_extent_info(trans, root,
6744 path->nodes[level]->start,
6745 path->nodes[level]->len,
6749 BUG_ON(wc->refs[level] == 0);
6751 if (level == root_item->drop_level)
6754 btrfs_tree_unlock(path->nodes[level]);
6755 WARN_ON(wc->refs[level] != 1);
6761 wc->shared_level = -1;
6762 wc->stage = DROP_REFERENCE;
6763 wc->update_ref = update_ref;
6765 wc->for_reloc = for_reloc;
6766 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6769 ret = walk_down_tree(trans, root, path, wc);
6775 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6782 BUG_ON(wc->stage != DROP_REFERENCE);
6786 if (wc->stage == DROP_REFERENCE) {
6788 btrfs_node_key(path->nodes[level],
6789 &root_item->drop_progress,
6790 path->slots[level]);
6791 root_item->drop_level = level;
6794 BUG_ON(wc->level == 0);
6795 if (btrfs_should_end_transaction(trans, tree_root)) {
6796 ret = btrfs_update_root(trans, tree_root,
6801 btrfs_end_transaction_throttle(trans, tree_root);
6802 trans = btrfs_start_transaction(tree_root, 0);
6803 BUG_ON(IS_ERR(trans));
6805 trans->block_rsv = block_rsv;
6808 btrfs_release_path(path);
6811 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6814 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6815 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6819 /* if we fail to delete the orphan item this time
6820 * around, it'll get picked up the next time.
6822 * The most common failure here is just -ENOENT.
6824 btrfs_del_orphan_item(trans, tree_root,
6825 root->root_key.objectid);
6829 if (root->in_radix) {
6830 btrfs_free_fs_root(tree_root->fs_info, root);
6832 free_extent_buffer(root->node);
6833 free_extent_buffer(root->commit_root);
6837 btrfs_end_transaction_throttle(trans, tree_root);
6839 btrfs_free_path(path);
6842 btrfs_std_error(root->fs_info, err);
6847 * drop subtree rooted at tree block 'node'.
6849 * NOTE: this function will unlock and release tree block 'node'
6850 * only used by relocation code
6852 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6853 struct btrfs_root *root,
6854 struct extent_buffer *node,
6855 struct extent_buffer *parent)
6857 struct btrfs_path *path;
6858 struct walk_control *wc;
6864 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6866 path = btrfs_alloc_path();
6870 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6872 btrfs_free_path(path);
6876 btrfs_assert_tree_locked(parent);
6877 parent_level = btrfs_header_level(parent);
6878 extent_buffer_get(parent);
6879 path->nodes[parent_level] = parent;
6880 path->slots[parent_level] = btrfs_header_nritems(parent);
6882 btrfs_assert_tree_locked(node);
6883 level = btrfs_header_level(node);
6884 path->nodes[level] = node;
6885 path->slots[level] = 0;
6886 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6888 wc->refs[parent_level] = 1;
6889 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6891 wc->shared_level = -1;
6892 wc->stage = DROP_REFERENCE;
6896 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6899 wret = walk_down_tree(trans, root, path, wc);
6905 wret = walk_up_tree(trans, root, path, wc, parent_level);
6913 btrfs_free_path(path);
6917 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6920 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6921 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6923 if (root->fs_info->balance_ctl) {
6924 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
6927 /* pick restriper's target profile and return */
6928 if (flags & BTRFS_BLOCK_GROUP_DATA &&
6929 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6930 tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
6931 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
6932 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6933 tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
6934 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
6935 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6936 tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
6940 /* extended -> chunk profile */
6941 tgt &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
6947 * we add in the count of missing devices because we want
6948 * to make sure that any RAID levels on a degraded FS
6949 * continue to be honored.
6951 num_devices = root->fs_info->fs_devices->rw_devices +
6952 root->fs_info->fs_devices->missing_devices;
6954 if (num_devices == 1) {
6955 stripped |= BTRFS_BLOCK_GROUP_DUP;
6956 stripped = flags & ~stripped;
6958 /* turn raid0 into single device chunks */
6959 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6962 /* turn mirroring into duplication */
6963 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6964 BTRFS_BLOCK_GROUP_RAID10))
6965 return stripped | BTRFS_BLOCK_GROUP_DUP;
6968 /* they already had raid on here, just return */
6969 if (flags & stripped)
6972 stripped |= BTRFS_BLOCK_GROUP_DUP;
6973 stripped = flags & ~stripped;
6975 /* switch duplicated blocks with raid1 */
6976 if (flags & BTRFS_BLOCK_GROUP_DUP)
6977 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6979 /* turn single device chunks into raid0 */
6980 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6985 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6987 struct btrfs_space_info *sinfo = cache->space_info;
6989 u64 min_allocable_bytes;
6994 * We need some metadata space and system metadata space for
6995 * allocating chunks in some corner cases until we force to set
6996 * it to be readonly.
6999 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7001 min_allocable_bytes = 1 * 1024 * 1024;
7003 min_allocable_bytes = 0;
7005 spin_lock(&sinfo->lock);
7006 spin_lock(&cache->lock);
7013 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7014 cache->bytes_super - btrfs_block_group_used(&cache->item);
7016 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7017 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7018 min_allocable_bytes <= sinfo->total_bytes) {
7019 sinfo->bytes_readonly += num_bytes;
7024 spin_unlock(&cache->lock);
7025 spin_unlock(&sinfo->lock);
7029 int btrfs_set_block_group_ro(struct btrfs_root *root,
7030 struct btrfs_block_group_cache *cache)
7033 struct btrfs_trans_handle *trans;
7039 trans = btrfs_join_transaction(root);
7040 BUG_ON(IS_ERR(trans));
7042 alloc_flags = update_block_group_flags(root, cache->flags);
7043 if (alloc_flags != cache->flags)
7044 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7047 ret = set_block_group_ro(cache, 0);
7050 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7051 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7055 ret = set_block_group_ro(cache, 0);
7057 btrfs_end_transaction(trans, root);
7061 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7062 struct btrfs_root *root, u64 type)
7064 u64 alloc_flags = get_alloc_profile(root, type);
7065 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7070 * helper to account the unused space of all the readonly block group in the
7071 * list. takes mirrors into account.
7073 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7075 struct btrfs_block_group_cache *block_group;
7079 list_for_each_entry(block_group, groups_list, list) {
7080 spin_lock(&block_group->lock);
7082 if (!block_group->ro) {
7083 spin_unlock(&block_group->lock);
7087 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7088 BTRFS_BLOCK_GROUP_RAID10 |
7089 BTRFS_BLOCK_GROUP_DUP))
7094 free_bytes += (block_group->key.offset -
7095 btrfs_block_group_used(&block_group->item)) *
7098 spin_unlock(&block_group->lock);
7105 * helper to account the unused space of all the readonly block group in the
7106 * space_info. takes mirrors into account.
7108 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7113 spin_lock(&sinfo->lock);
7115 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7116 if (!list_empty(&sinfo->block_groups[i]))
7117 free_bytes += __btrfs_get_ro_block_group_free_space(
7118 &sinfo->block_groups[i]);
7120 spin_unlock(&sinfo->lock);
7125 int btrfs_set_block_group_rw(struct btrfs_root *root,
7126 struct btrfs_block_group_cache *cache)
7128 struct btrfs_space_info *sinfo = cache->space_info;
7133 spin_lock(&sinfo->lock);
7134 spin_lock(&cache->lock);
7135 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7136 cache->bytes_super - btrfs_block_group_used(&cache->item);
7137 sinfo->bytes_readonly -= num_bytes;
7139 spin_unlock(&cache->lock);
7140 spin_unlock(&sinfo->lock);
7145 * checks to see if its even possible to relocate this block group.
7147 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7148 * ok to go ahead and try.
7150 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7152 struct btrfs_block_group_cache *block_group;
7153 struct btrfs_space_info *space_info;
7154 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7155 struct btrfs_device *device;
7163 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7165 /* odd, couldn't find the block group, leave it alone */
7169 min_free = btrfs_block_group_used(&block_group->item);
7171 /* no bytes used, we're good */
7175 space_info = block_group->space_info;
7176 spin_lock(&space_info->lock);
7178 full = space_info->full;
7181 * if this is the last block group we have in this space, we can't
7182 * relocate it unless we're able to allocate a new chunk below.
7184 * Otherwise, we need to make sure we have room in the space to handle
7185 * all of the extents from this block group. If we can, we're good
7187 if ((space_info->total_bytes != block_group->key.offset) &&
7188 (space_info->bytes_used + space_info->bytes_reserved +
7189 space_info->bytes_pinned + space_info->bytes_readonly +
7190 min_free < space_info->total_bytes)) {
7191 spin_unlock(&space_info->lock);
7194 spin_unlock(&space_info->lock);
7197 * ok we don't have enough space, but maybe we have free space on our
7198 * devices to allocate new chunks for relocation, so loop through our
7199 * alloc devices and guess if we have enough space. However, if we
7200 * were marked as full, then we know there aren't enough chunks, and we
7215 index = get_block_group_index(block_group);
7220 } else if (index == 1) {
7222 } else if (index == 2) {
7225 } else if (index == 3) {
7226 dev_min = fs_devices->rw_devices;
7227 do_div(min_free, dev_min);
7230 mutex_lock(&root->fs_info->chunk_mutex);
7231 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7235 * check to make sure we can actually find a chunk with enough
7236 * space to fit our block group in.
7238 if (device->total_bytes > device->bytes_used + min_free) {
7239 ret = find_free_dev_extent(device, min_free,
7244 if (dev_nr >= dev_min)
7250 mutex_unlock(&root->fs_info->chunk_mutex);
7252 btrfs_put_block_group(block_group);
7256 static int find_first_block_group(struct btrfs_root *root,
7257 struct btrfs_path *path, struct btrfs_key *key)
7260 struct btrfs_key found_key;
7261 struct extent_buffer *leaf;
7264 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7269 slot = path->slots[0];
7270 leaf = path->nodes[0];
7271 if (slot >= btrfs_header_nritems(leaf)) {
7272 ret = btrfs_next_leaf(root, path);
7279 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7281 if (found_key.objectid >= key->objectid &&
7282 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7292 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7294 struct btrfs_block_group_cache *block_group;
7298 struct inode *inode;
7300 block_group = btrfs_lookup_first_block_group(info, last);
7301 while (block_group) {
7302 spin_lock(&block_group->lock);
7303 if (block_group->iref)
7305 spin_unlock(&block_group->lock);
7306 block_group = next_block_group(info->tree_root,
7316 inode = block_group->inode;
7317 block_group->iref = 0;
7318 block_group->inode = NULL;
7319 spin_unlock(&block_group->lock);
7321 last = block_group->key.objectid + block_group->key.offset;
7322 btrfs_put_block_group(block_group);
7326 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7328 struct btrfs_block_group_cache *block_group;
7329 struct btrfs_space_info *space_info;
7330 struct btrfs_caching_control *caching_ctl;
7333 down_write(&info->extent_commit_sem);
7334 while (!list_empty(&info->caching_block_groups)) {
7335 caching_ctl = list_entry(info->caching_block_groups.next,
7336 struct btrfs_caching_control, list);
7337 list_del(&caching_ctl->list);
7338 put_caching_control(caching_ctl);
7340 up_write(&info->extent_commit_sem);
7342 spin_lock(&info->block_group_cache_lock);
7343 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7344 block_group = rb_entry(n, struct btrfs_block_group_cache,
7346 rb_erase(&block_group->cache_node,
7347 &info->block_group_cache_tree);
7348 spin_unlock(&info->block_group_cache_lock);
7350 down_write(&block_group->space_info->groups_sem);
7351 list_del(&block_group->list);
7352 up_write(&block_group->space_info->groups_sem);
7354 if (block_group->cached == BTRFS_CACHE_STARTED)
7355 wait_block_group_cache_done(block_group);
7358 * We haven't cached this block group, which means we could
7359 * possibly have excluded extents on this block group.
7361 if (block_group->cached == BTRFS_CACHE_NO)
7362 free_excluded_extents(info->extent_root, block_group);
7364 btrfs_remove_free_space_cache(block_group);
7365 btrfs_put_block_group(block_group);
7367 spin_lock(&info->block_group_cache_lock);
7369 spin_unlock(&info->block_group_cache_lock);
7371 /* now that all the block groups are freed, go through and
7372 * free all the space_info structs. This is only called during
7373 * the final stages of unmount, and so we know nobody is
7374 * using them. We call synchronize_rcu() once before we start,
7375 * just to be on the safe side.
7379 release_global_block_rsv(info);
7381 while(!list_empty(&info->space_info)) {
7382 space_info = list_entry(info->space_info.next,
7383 struct btrfs_space_info,
7385 if (space_info->bytes_pinned > 0 ||
7386 space_info->bytes_reserved > 0 ||
7387 space_info->bytes_may_use > 0) {
7389 dump_space_info(space_info, 0, 0);
7391 list_del(&space_info->list);
7397 static void __link_block_group(struct btrfs_space_info *space_info,
7398 struct btrfs_block_group_cache *cache)
7400 int index = get_block_group_index(cache);
7402 down_write(&space_info->groups_sem);
7403 list_add_tail(&cache->list, &space_info->block_groups[index]);
7404 up_write(&space_info->groups_sem);
7407 int btrfs_read_block_groups(struct btrfs_root *root)
7409 struct btrfs_path *path;
7411 struct btrfs_block_group_cache *cache;
7412 struct btrfs_fs_info *info = root->fs_info;
7413 struct btrfs_space_info *space_info;
7414 struct btrfs_key key;
7415 struct btrfs_key found_key;
7416 struct extent_buffer *leaf;
7420 root = info->extent_root;
7423 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7424 path = btrfs_alloc_path();
7429 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7430 if (btrfs_test_opt(root, SPACE_CACHE) &&
7431 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7433 if (btrfs_test_opt(root, CLEAR_CACHE))
7437 ret = find_first_block_group(root, path, &key);
7442 leaf = path->nodes[0];
7443 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7444 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7449 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7451 if (!cache->free_space_ctl) {
7457 atomic_set(&cache->count, 1);
7458 spin_lock_init(&cache->lock);
7459 cache->fs_info = info;
7460 INIT_LIST_HEAD(&cache->list);
7461 INIT_LIST_HEAD(&cache->cluster_list);
7464 cache->disk_cache_state = BTRFS_DC_CLEAR;
7466 read_extent_buffer(leaf, &cache->item,
7467 btrfs_item_ptr_offset(leaf, path->slots[0]),
7468 sizeof(cache->item));
7469 memcpy(&cache->key, &found_key, sizeof(found_key));
7471 key.objectid = found_key.objectid + found_key.offset;
7472 btrfs_release_path(path);
7473 cache->flags = btrfs_block_group_flags(&cache->item);
7474 cache->sectorsize = root->sectorsize;
7476 btrfs_init_free_space_ctl(cache);
7479 * We need to exclude the super stripes now so that the space
7480 * info has super bytes accounted for, otherwise we'll think
7481 * we have more space than we actually do.
7483 exclude_super_stripes(root, cache);
7486 * check for two cases, either we are full, and therefore
7487 * don't need to bother with the caching work since we won't
7488 * find any space, or we are empty, and we can just add all
7489 * the space in and be done with it. This saves us _alot_ of
7490 * time, particularly in the full case.
7492 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7493 cache->last_byte_to_unpin = (u64)-1;
7494 cache->cached = BTRFS_CACHE_FINISHED;
7495 free_excluded_extents(root, cache);
7496 } else if (btrfs_block_group_used(&cache->item) == 0) {
7497 cache->last_byte_to_unpin = (u64)-1;
7498 cache->cached = BTRFS_CACHE_FINISHED;
7499 add_new_free_space(cache, root->fs_info,
7501 found_key.objectid +
7503 free_excluded_extents(root, cache);
7506 ret = update_space_info(info, cache->flags, found_key.offset,
7507 btrfs_block_group_used(&cache->item),
7510 cache->space_info = space_info;
7511 spin_lock(&cache->space_info->lock);
7512 cache->space_info->bytes_readonly += cache->bytes_super;
7513 spin_unlock(&cache->space_info->lock);
7515 __link_block_group(space_info, cache);
7517 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7520 set_avail_alloc_bits(root->fs_info, cache->flags);
7521 if (btrfs_chunk_readonly(root, cache->key.objectid))
7522 set_block_group_ro(cache, 1);
7525 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7526 if (!(get_alloc_profile(root, space_info->flags) &
7527 (BTRFS_BLOCK_GROUP_RAID10 |
7528 BTRFS_BLOCK_GROUP_RAID1 |
7529 BTRFS_BLOCK_GROUP_DUP)))
7532 * avoid allocating from un-mirrored block group if there are
7533 * mirrored block groups.
7535 list_for_each_entry(cache, &space_info->block_groups[3], list)
7536 set_block_group_ro(cache, 1);
7537 list_for_each_entry(cache, &space_info->block_groups[4], list)
7538 set_block_group_ro(cache, 1);
7541 init_global_block_rsv(info);
7544 btrfs_free_path(path);
7548 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7549 struct btrfs_root *root, u64 bytes_used,
7550 u64 type, u64 chunk_objectid, u64 chunk_offset,
7554 struct btrfs_root *extent_root;
7555 struct btrfs_block_group_cache *cache;
7557 extent_root = root->fs_info->extent_root;
7559 root->fs_info->last_trans_log_full_commit = trans->transid;
7561 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7564 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7566 if (!cache->free_space_ctl) {
7571 cache->key.objectid = chunk_offset;
7572 cache->key.offset = size;
7573 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7574 cache->sectorsize = root->sectorsize;
7575 cache->fs_info = root->fs_info;
7577 atomic_set(&cache->count, 1);
7578 spin_lock_init(&cache->lock);
7579 INIT_LIST_HEAD(&cache->list);
7580 INIT_LIST_HEAD(&cache->cluster_list);
7582 btrfs_init_free_space_ctl(cache);
7584 btrfs_set_block_group_used(&cache->item, bytes_used);
7585 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7586 cache->flags = type;
7587 btrfs_set_block_group_flags(&cache->item, type);
7589 cache->last_byte_to_unpin = (u64)-1;
7590 cache->cached = BTRFS_CACHE_FINISHED;
7591 exclude_super_stripes(root, cache);
7593 add_new_free_space(cache, root->fs_info, chunk_offset,
7594 chunk_offset + size);
7596 free_excluded_extents(root, cache);
7598 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7599 &cache->space_info);
7601 update_global_block_rsv(root->fs_info);
7603 spin_lock(&cache->space_info->lock);
7604 cache->space_info->bytes_readonly += cache->bytes_super;
7605 spin_unlock(&cache->space_info->lock);
7607 __link_block_group(cache->space_info, cache);
7609 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7612 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7613 sizeof(cache->item));
7616 set_avail_alloc_bits(extent_root->fs_info, type);
7621 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7623 u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
7625 /* chunk -> extended profile */
7626 if (extra_flags == 0)
7627 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
7629 if (flags & BTRFS_BLOCK_GROUP_DATA)
7630 fs_info->avail_data_alloc_bits &= ~extra_flags;
7631 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7632 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7633 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7634 fs_info->avail_system_alloc_bits &= ~extra_flags;
7637 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7638 struct btrfs_root *root, u64 group_start)
7640 struct btrfs_path *path;
7641 struct btrfs_block_group_cache *block_group;
7642 struct btrfs_free_cluster *cluster;
7643 struct btrfs_root *tree_root = root->fs_info->tree_root;
7644 struct btrfs_key key;
7645 struct inode *inode;
7650 root = root->fs_info->extent_root;
7652 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7653 BUG_ON(!block_group);
7654 BUG_ON(!block_group->ro);
7657 * Free the reserved super bytes from this block group before
7660 free_excluded_extents(root, block_group);
7662 memcpy(&key, &block_group->key, sizeof(key));
7663 index = get_block_group_index(block_group);
7664 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7665 BTRFS_BLOCK_GROUP_RAID1 |
7666 BTRFS_BLOCK_GROUP_RAID10))
7671 /* make sure this block group isn't part of an allocation cluster */
7672 cluster = &root->fs_info->data_alloc_cluster;
7673 spin_lock(&cluster->refill_lock);
7674 btrfs_return_cluster_to_free_space(block_group, cluster);
7675 spin_unlock(&cluster->refill_lock);
7678 * make sure this block group isn't part of a metadata
7679 * allocation cluster
7681 cluster = &root->fs_info->meta_alloc_cluster;
7682 spin_lock(&cluster->refill_lock);
7683 btrfs_return_cluster_to_free_space(block_group, cluster);
7684 spin_unlock(&cluster->refill_lock);
7686 path = btrfs_alloc_path();
7692 inode = lookup_free_space_inode(tree_root, block_group, path);
7693 if (!IS_ERR(inode)) {
7694 ret = btrfs_orphan_add(trans, inode);
7697 /* One for the block groups ref */
7698 spin_lock(&block_group->lock);
7699 if (block_group->iref) {
7700 block_group->iref = 0;
7701 block_group->inode = NULL;
7702 spin_unlock(&block_group->lock);
7705 spin_unlock(&block_group->lock);
7707 /* One for our lookup ref */
7708 btrfs_add_delayed_iput(inode);
7711 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7712 key.offset = block_group->key.objectid;
7715 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7719 btrfs_release_path(path);
7721 ret = btrfs_del_item(trans, tree_root, path);
7724 btrfs_release_path(path);
7727 spin_lock(&root->fs_info->block_group_cache_lock);
7728 rb_erase(&block_group->cache_node,
7729 &root->fs_info->block_group_cache_tree);
7730 spin_unlock(&root->fs_info->block_group_cache_lock);
7732 down_write(&block_group->space_info->groups_sem);
7734 * we must use list_del_init so people can check to see if they
7735 * are still on the list after taking the semaphore
7737 list_del_init(&block_group->list);
7738 if (list_empty(&block_group->space_info->block_groups[index]))
7739 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7740 up_write(&block_group->space_info->groups_sem);
7742 if (block_group->cached == BTRFS_CACHE_STARTED)
7743 wait_block_group_cache_done(block_group);
7745 btrfs_remove_free_space_cache(block_group);
7747 spin_lock(&block_group->space_info->lock);
7748 block_group->space_info->total_bytes -= block_group->key.offset;
7749 block_group->space_info->bytes_readonly -= block_group->key.offset;
7750 block_group->space_info->disk_total -= block_group->key.offset * factor;
7751 spin_unlock(&block_group->space_info->lock);
7753 memcpy(&key, &block_group->key, sizeof(key));
7755 btrfs_clear_space_info_full(root->fs_info);
7757 btrfs_put_block_group(block_group);
7758 btrfs_put_block_group(block_group);
7760 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7766 ret = btrfs_del_item(trans, root, path);
7768 btrfs_free_path(path);
7772 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7774 struct btrfs_space_info *space_info;
7775 struct btrfs_super_block *disk_super;
7781 disk_super = fs_info->super_copy;
7782 if (!btrfs_super_root(disk_super))
7785 features = btrfs_super_incompat_flags(disk_super);
7786 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7789 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7790 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7795 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7796 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7798 flags = BTRFS_BLOCK_GROUP_METADATA;
7799 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7803 flags = BTRFS_BLOCK_GROUP_DATA;
7804 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7810 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7812 return unpin_extent_range(root, start, end);
7815 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7816 u64 num_bytes, u64 *actual_bytes)
7818 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7821 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7823 struct btrfs_fs_info *fs_info = root->fs_info;
7824 struct btrfs_block_group_cache *cache = NULL;
7831 cache = btrfs_lookup_block_group(fs_info, range->start);
7834 if (cache->key.objectid >= (range->start + range->len)) {
7835 btrfs_put_block_group(cache);
7839 start = max(range->start, cache->key.objectid);
7840 end = min(range->start + range->len,
7841 cache->key.objectid + cache->key.offset);
7843 if (end - start >= range->minlen) {
7844 if (!block_group_cache_done(cache)) {
7845 ret = cache_block_group(cache, NULL, root, 0);
7847 wait_block_group_cache_done(cache);
7849 ret = btrfs_trim_block_group(cache,
7855 trimmed += group_trimmed;
7857 btrfs_put_block_group(cache);
7862 cache = next_block_group(fs_info->tree_root, cache);
7865 range->len = trimmed;